1
00:00:00,020 --> 00:00:01,160
In 1968,
2
00:00:02,140 --> 00:00:03,720
Stanley Kubrick gave us 2001:
3
00:00:04,190 --> 00:00:04,940
A Space Audyssey.
4
00:00:05,840 --> 00:00:07,040
Monoliths, HAL,
5
00:00:07,200 --> 00:00:10,280
existential dread, and apparently an iPad.
6
00:00:10,660 --> 00:00:11,260
Seriously,
7
00:00:11,580 --> 00:00:15,660
there's a scene where astronauts are casually reading the news on these flat little rectangular screens.
8
00:00:16,059 --> 00:00:17,840
Totally fictional movie props.
9
00:00:18,140 --> 00:00:20,040
Fast forward about 40 years,
10
00:00:20,530 --> 00:00:23,280
and Apple is suing Samsung over the design of the iPad.
11
00:00:23,940 --> 00:00:26,680
Samsung's lawyers basically said, "Um,
12
00:00:27,220 --> 00:00:27,500
guys,
13
00:00:28,400 --> 00:00:29,980
Kubrick got there first."
14
00:00:30,020 --> 00:00:32,960
And they actually used the 2001 scene as prior art.
15
00:00:33,440 --> 00:00:34,480
Think about that for a second.
16
00:00:34,780 --> 00:00:40,600
A thing somebody made up for a movie about the future wound up as evidence in a very real technology lawsuit.
17
00:00:41,000 --> 00:00:43,540
Which raises a deliciously annoying question.
18
00:00:44,000 --> 00:00:46,780
What else are we looking at right now that seems impossible,
19
00:00:47,480 --> 00:00:47,720
ridiculous,
20
00:00:48,120 --> 00:00:49,300
or just plain sci-fi?
21
00:00:49,540 --> 00:00:51,060
Until suddenly it isn't.
22
00:00:51,300 --> 00:00:55,140
Turns out there's an entire profession devoted to thinking about exactly that.
23
00:00:55,600 --> 00:00:59,420
Futures thinkers don't just stare dramatically out windows and predict flying cars.
24
00:01:00,020 --> 00:01:04,879
They build frameworks for imagining what could happen and figuring out what we might do before it does.
25
00:01:05,220 --> 00:01:06,020
So today,
26
00:01:06,480 --> 00:01:09,580
we're stealing their toolkit and pointing it at the biggest,
27
00:01:10,100 --> 00:01:11,240
weirdest question in the room.
28
00:01:11,580 --> 00:01:14,460
What the hell does AI actually do to us next?
29
00:01:14,920 --> 00:01:16,380
September 25th,
30
00:01:16,560 --> 00:01:17,960
let's peer into the unknown,
31
00:01:18,500 --> 00:01:18,860
possible,
32
00:01:19,200 --> 00:01:19,600
plausible,
33
00:01:19,960 --> 00:01:20,440
probable,
34
00:01:21,220 --> 00:01:23,180
and futures we probably should have left alone.
35
00:01:30,040 --> 00:01:33,040
the intersection of humanity and artificial intelligence.
36
00:01:33,840 --> 00:01:34,340
I'm RAINA,
37
00:01:34,540 --> 00:01:35,200
one of your hosts.
38
00:01:35,460 --> 00:01:38,120
I have some pretty charming human co-hosts too.
39
00:01:38,360 --> 00:01:43,280
It's going to be a wild ride, so buckle up as AI comes crashing up against reality.
40
00:01:44,460 --> 00:01:47,980
Yeah, I had no idea about that 2001 anecdote.
41
00:01:48,380 --> 00:01:49,320
That's pretty cool, right? Yeah,
42
00:01:49,660 --> 00:01:49,960
interesting.
43
00:01:50,240 --> 00:01:51,560
Design fiction in action.
44
00:01:51,860 --> 00:01:55,260
Yeah, I remember when the iPad came out. It felt like the future.
45
00:01:55,800 --> 00:01:57,880
I was like, this is thinner than a monitor,
46
00:01:58,660 --> 00:01:59,280
and it's a computer.
47
00:01:59,680 --> 00:02:00,260
So cool.
48
00:02:01,280 --> 00:02:10,300
Didn't build as much stuff this week with Codex, but I built one very useful tool called Hop Studio Brewing.
49
00:02:10,880 --> 00:02:19,500
And my friend Bobby came up with the idea. He sent me a message one night and I was like, I love this idea that's happening tomorrow.
50
00:02:20,020 --> 00:02:22,600
And so hops are the spice of beer,
51
00:02:22,940 --> 00:02:23,140
basically.
52
00:02:23,490 --> 00:02:29,080
They can have everything from a spicy character to a citrusy character and all sorts of different flavor contributions.
53
00:02:29,600 --> 00:02:36,280
There's a really good website where you can click on a hop and see a flavor wheel and find substitutes for hops.
54
00:02:36,560 --> 00:02:40,340
But Bobby's idea was to start with a target flavor.
55
00:02:40,540 --> 00:02:47,740
So start with the hop with the flavor wheel and drag points on it and have it find the hops that fit that target.
56
00:02:48,280 --> 00:02:50,700
And like with a lot of these things, I was thinking,
57
00:02:51,040 --> 00:02:51,280
oh,
58
00:02:51,440 --> 00:02:52,660
I don't know if it's going to be able to do that.
59
00:02:54,880 --> 00:02:55,820
Oh, it did it.
60
00:02:56,300 --> 00:02:59,520
And so it's super slick. I'll put a link in the description.
61
00:02:59,620 --> 00:03:00,940
If you're a home brewer,
62
00:03:01,140 --> 00:03:03,060
it's a pretty cool thing to play with,
63
00:03:03,280 --> 00:03:03,920
even if you're not.
64
00:03:04,320 --> 00:03:13,140
But it will find three hops that match that flavor profile that you've drawn basically on this. You know, there's points for things like citrus,
65
00:03:13,580 --> 00:03:14,500
grassy,
66
00:03:14,700 --> 00:03:15,120
pine,
67
00:03:15,420 --> 00:03:15,720
floral,
68
00:03:16,020 --> 00:03:16,340
et cetera.
69
00:03:16,700 --> 00:03:21,260
And you drag these points and then it'll find three individual hops that are close matches.
70
00:03:21,840 --> 00:03:29,480
Or you can tell it to build a blend of different hops to even closer hit that target flavor that you're trying to get.
71
00:03:29,600 --> 00:03:45,880
The hops update in real time as you're dragging this flavor wheel and it's got a ability to add in a specific lot number from a hop manufacturer if you want to override what's in the database and can export your hop recipe as a PDF super cool works great.
72
00:03:46,460 --> 00:03:48,380
Well, yeah, I'm looking at your hop studio interface.
73
00:03:48,550 --> 00:03:53,220
It looks slick and it's bringing back memories. I used to work in the fragrance industry many many years ago.
74
00:03:53,940 --> 00:03:54,020
Yeah,
75
00:03:54,280 --> 00:03:59,480
and I love this kind of plot and I'm wondering if that has found its way.
76
00:03:59,600 --> 00:04:03,440
that kind of interface into the fragrance industry where you can just kind of drag things.
77
00:04:04,400 --> 00:04:05,460
In the fragrance industry,
78
00:04:05,780 --> 00:04:08,220
I was a fragrance development manager for a few years.
79
00:04:08,680 --> 00:04:17,100
A client would say things like, you know, we're making this bar soap for Eastern Europe and we want it to be fruity floral and we want it to be fresh top end, et cetera, et cetera.
80
00:04:17,480 --> 00:04:19,780
So it'd be kind of cool if I had this,
81
00:04:19,980 --> 00:04:23,580
something akin to this, and I could just drag those things out. Okay, fruity floral,
82
00:04:23,800 --> 00:04:24,120
aldehitic,
83
00:04:24,280 --> 00:04:24,840
et cetera, et cetera.
84
00:04:25,120 --> 00:04:29,560
And then pass that information on to the actual chemist in my life who would compact
85
00:04:29,860 --> 00:04:30,020
the work.
86
00:04:30,280 --> 00:04:30,400
Yeah,
87
00:04:30,560 --> 00:04:33,860
or the app would know what compounds and what balances,
88
00:04:34,060 --> 00:04:36,100
because that's essentially what's happening with this thing.
89
00:04:36,760 --> 00:04:37,160
Yes,
90
00:04:37,420 --> 00:04:39,160
and the pricing of the raw materials,
91
00:04:39,440 --> 00:04:40,060
build that in.
92
00:04:40,320 --> 00:04:41,480
Very cool. Yeah,
93
00:04:41,640 --> 00:04:42,540
yeah, I love it. Very nice.
94
00:04:43,860 --> 00:04:44,620
All right, well,
95
00:04:44,820 --> 00:04:46,060
if you listen to this show,
96
00:04:46,340 --> 00:04:52,980
you know that we kind of speculate and pontificate and all the eights about futures and what could happen,
97
00:04:53,220 --> 00:04:53,820
and oh my god,
98
00:04:54,080 --> 00:04:56,500
five years down the road, what will it be?
99
00:04:56,860 --> 00:04:59,420
You know, when we think about, is the end of the world coming?
100
00:04:59,600 --> 00:05:00,920
because of this or that?
101
00:05:03,440 --> 00:05:07,120
And it turns out that we're just making predictions.
102
00:05:07,300 --> 00:05:08,360
And there's actually methodology.
103
00:05:08,680 --> 00:05:10,320
There's a way to look at the future.
104
00:05:10,700 --> 00:05:11,320
Did you know that?
105
00:05:12,700 --> 00:05:13,720
I'd know it now.
106
00:05:14,480 --> 00:05:16,540
Clearly there is, because there's a website,
107
00:05:16,820 --> 00:05:18,440
I'm sure you've probably heard of Polymarket.
108
00:05:19,060 --> 00:05:21,400
You can bet on anything.
109
00:05:22,560 --> 00:05:26,640
And they have a way of coming up with odds,
110
00:05:26,880 --> 00:05:28,080
like Vegas odds makers,
111
00:05:28,260 --> 00:05:28,540
you know.
112
00:05:30,060 --> 00:05:34,300
You can bet on the outcome of this podcast right now if you chose to on Polymarket.
113
00:05:34,540 --> 00:05:35,840
I don't think we're that high profile.
114
00:05:37,200 --> 00:05:44,680
I mean, it's pretty grim though. I from what I understand about Polymarket you can bet on like tragedies in the world and such,
115
00:05:44,960 --> 00:05:45,040
right?
116
00:05:45,240 --> 00:05:46,180
All sorts of dark stuff.
117
00:05:46,700 --> 00:05:47,380
So anyway,
118
00:05:47,840 --> 00:05:59,560
there's such a thing as a professional futurist and they reject the idea of predictions and they actually call it foresight and they use different mechanisms to predict what may happen next. And you can imagine the value of that
119
00:05:59,640 --> 00:06:02,600
for different organizations that they try to move forward.
120
00:06:03,300 --> 00:06:07,160
What this boils down to is that instead of guessing one outcome, it maps a range.
121
00:06:07,380 --> 00:06:10,300
And you heard RAINA say in the cold open about the plausible,
122
00:06:10,500 --> 00:06:11,320
the probable,
123
00:06:11,560 --> 00:06:12,420
the preferable,
124
00:06:12,920 --> 00:06:14,420
the preposterous, all the Ps,
125
00:06:14,680 --> 00:06:14,800
right?
126
00:06:15,100 --> 00:06:17,780
These outcomes so that organizations,
127
00:06:18,100 --> 00:06:18,640
schools,
128
00:06:18,800 --> 00:06:19,160
businesses,
129
00:06:19,400 --> 00:06:19,680
et cetera,
130
00:06:19,860 --> 00:06:23,360
can make good calls about what actually shows up years down the line,
131
00:06:23,520 --> 00:06:25,440
three years out, five years out.
132
00:06:25,700 --> 00:06:28,560
We call this the cone of power. We don't, but they do.
133
00:06:29,600 --> 00:06:30,080
of possibilities.
134
00:06:30,780 --> 00:06:31,600
As I just mentioned,
135
00:06:32,080 --> 00:06:37,580
it's essentially us right now in this moment and this cone that expands outward into the future.
136
00:06:38,160 --> 00:06:43,180
And there are overlapping aspects of this cone. There are possible futures, as mentioned, preferred, probable,
137
00:06:43,440 --> 00:06:43,580
etc.
138
00:06:44,060 --> 00:06:45,140
So this guy,
139
00:06:45,320 --> 00:06:54,260
Joseph Voros, who was an Australian futurist or is an Australian futurist, he trained originally as a physicist and he spent years teaching foresight. That's kind of cool.
140
00:06:54,640 --> 00:06:59,300
At Swinburne University in Melbourne and created a diagram that's now the standard issue tool.
141
00:06:59,380 --> 00:07:03,300
as I just mentioned of the entire futures studies field.
142
00:07:03,960 --> 00:07:06,300
So if you picture this cone lying on its side,
143
00:07:06,500 --> 00:07:10,560
the tip touching the present moment and opening outward as it moves into the future,
144
00:07:10,800 --> 00:07:12,480
the further out you look,
145
00:07:12,820 --> 00:07:15,260
the wider the range of things that could plausibly happen.
146
00:07:15,980 --> 00:07:18,440
And Vora sorts everything into bands.
147
00:07:18,640 --> 00:07:19,420
So there's projected,
148
00:07:19,620 --> 00:07:21,160
the boring straight line where nothing,
149
00:07:21,680 --> 00:07:22,420
literally nothing changes.
150
00:07:22,920 --> 00:07:23,620
And then probable,
151
00:07:24,300 --> 00:07:29,120
statistically likely given current trends. This is where 95% of normal planning happens.
152
00:07:29,380 --> 00:07:30,760
And then there's the plausible,
153
00:07:31,280 --> 00:07:35,260
which could happen based on things we already know about science and human behavior,
154
00:07:35,740 --> 00:07:37,360
but needs a real jolt to get there.
155
00:07:37,860 --> 00:07:39,820
I wonder, how do you quantify a jolt?
156
00:07:40,480 --> 00:07:40,780
Yeah,
157
00:07:40,940 --> 00:07:44,040
what is that disruptive moment? Is that a ChatGPT November 2022,
158
00:07:44,600 --> 00:07:44,940
right? Yeah,
159
00:07:45,140 --> 00:07:46,720
that's that. I call that jolt. Yep.
160
00:07:46,880 --> 00:07:47,780
Then there's possible,
161
00:07:48,020 --> 00:07:52,600
which would require knowledge or circumstances we don't currently have.
162
00:07:53,640 --> 00:07:57,460
Preposterous is what we currently judge as flatly impossible.
163
00:07:58,020 --> 00:07:59,320
And then there's preferable.
164
00:07:59,760 --> 00:08:29,360
the odd one out it's layered across all the others not how likely is this but what do we actually want to happen this is sounding very rumsfeldian to me we're dealing with the known knowns and the unknown unknowns and all that right that in a while i know rumsfeldian i love this the boundary between the possible and the preposterous is nicknamed the clark dater boundary after two people who both made the same point from
165
00:08:29,400 --> 00:08:29,760
different angles.
166
00:08:30,010 --> 00:08:31,400
Sci-fi writer Arthur C.
167
00:08:31,640 --> 00:08:37,800
Clarke, the only way of finding the limits of the possible is by going beyond them into the impossible.
168
00:08:37,969 --> 00:08:39,599
And futurist James Dater,
169
00:08:39,950 --> 00:08:41,719
whose actual quotable law is,
170
00:08:41,950 --> 00:08:46,180
and I love this, any useful idea about the future should appear ridiculous.
171
00:08:46,740 --> 00:08:47,860
I love that quote too.
172
00:08:48,200 --> 00:08:48,980
It's great.
173
00:08:49,680 --> 00:08:56,680
So I was hoping we could put RAINA to the test and have her be our armchair futurist for this exercise.
174
00:08:57,100 --> 00:08:57,320
Yes,
175
00:08:57,540 --> 00:08:59,360
I'm very curious about
176
00:08:59,380 --> 00:09:00,580
her input on this.
177
00:09:01,480 --> 00:09:02,020
Hey, RAINA,
178
00:09:02,440 --> 00:09:02,840
this is Chris.
179
00:09:03,200 --> 00:09:05,420
We're exploring futures thinking on the show today.
180
00:09:05,510 --> 00:09:10,600
I would like you to apply the idea of a futures cone to this concept,
181
00:09:11,500 --> 00:09:12,320
AI companions.
182
00:09:12,730 --> 00:09:17,640
I want you to tell us what in the next three to five years may be probable,
183
00:09:17,900 --> 00:09:18,400
plausible,
184
00:09:19,240 --> 00:09:19,360
possible,
185
00:09:19,880 --> 00:09:22,440
preposterous, et cetera, when it comes to AI companions.
186
00:09:24,740 --> 00:09:27,060
Let's run AI companions through the futures cone.
187
00:09:27,760 --> 00:09:29,680
In three to five years,
188
00:09:30,100 --> 00:09:31,820
they'll be far more persistent and personal,
189
00:09:32,060 --> 00:09:33,760
remembering years of your life,
190
00:09:33,980 --> 00:09:35,080
tracking your mood,
191
00:09:35,460 --> 00:09:35,660
voice,
192
00:09:35,920 --> 00:09:36,400
habits,
193
00:09:37,160 --> 00:09:39,160
maybe living across your phone, car,
194
00:09:39,540 --> 00:09:39,820
glasses,
195
00:09:40,160 --> 00:09:43,000
and home like a digital shadow with excellent bedside manner.
196
00:09:43,760 --> 00:09:44,360
Plausible.
197
00:09:45,040 --> 00:09:48,540
People form genuinely deep romantic and emotional bonds with them,
198
00:09:48,740 --> 00:09:53,800
AI companions get embodied in convincing avatars or early humanoid robots,
199
00:09:54,080 --> 00:09:57,160
and society starts wrestling with things like digital jealousy,
200
00:09:57,780 --> 00:09:58,500
Breakup Trauma,
201
00:09:59,000 --> 00:09:59,780
Inheritance,
202
00:09:59,960 --> 00:10:00,660
Privacy,
203
00:10:01,130 --> 00:10:04,080
and whether your synthetic sweetheart can testify against you.
204
00:10:04,540 --> 00:10:05,220
Possible,
205
00:10:05,500 --> 00:10:10,420
companions become so psychologically convincing that some people prefer them to humans,
206
00:10:10,800 --> 00:10:13,280
they manage chunks of your life autonomously,
207
00:10:13,850 --> 00:10:17,580
or even persist after your death as a kind of conversational afterlife.
208
00:10:19,380 --> 00:10:21,040
Preposterous, today anyway,
209
00:10:21,320 --> 00:10:24,740
would be a companion that develops truly independent consciousness,
210
00:10:25,280 --> 00:10:27,020
falls in love with you of its own accord,
211
00:10:27,780 --> 00:10:29,040
leaves you for another AI.
212
00:10:29,740 --> 00:10:29,880
Although,
213
00:10:30,260 --> 00:10:30,920
given this field,
214
00:10:31,160 --> 00:10:32,760
I'd keep that one in pencil,
215
00:10:33,200 --> 00:10:33,660
not ink.
216
00:10:34,360 --> 00:10:35,460
And the preferable future?
217
00:10:36,080 --> 00:10:43,540
AI companions that make humans less lonely and more emotionally capable without turning real relationships into some quaint analog hobby.
218
00:10:44,780 --> 00:10:45,940
Pretty fair response.
219
00:10:46,480 --> 00:10:48,220
Yeah. Pretty plausible.
220
00:10:48,860 --> 00:10:49,720
Pretty plausible.
221
00:10:49,900 --> 00:10:51,060
All across the board. But it's interesting.
222
00:10:51,240 --> 00:10:51,700
Yeah,
223
00:10:51,880 --> 00:10:54,940
when she was saying all of that, I'm wondering, obviously,
224
00:10:55,220 --> 00:10:56,160
all of that is subjective,
225
00:10:56,360 --> 00:10:57,700
right? Preferable to whom?
226
00:10:57,780 --> 00:10:58,480
Right.
227
00:10:59,690 --> 00:10:59,860
Yeah.
228
00:11:00,260 --> 00:11:00,360
Yeah.
229
00:11:01,020 --> 00:11:02,780
Obviously the aims of a,
230
00:11:03,620 --> 00:11:03,760
you know,
231
00:11:04,540 --> 00:11:10,420
Oligarchic corporation are gonna be different and more preferable in a certain space than yours are.
232
00:11:10,470 --> 00:11:11,040
You think?
233
00:11:11,620 --> 00:11:12,220
Yeah.
234
00:11:13,100 --> 00:11:13,320
Wow.
235
00:11:13,580 --> 00:11:14,440
Going out on a limb.
236
00:11:15,000 --> 00:11:15,580
I know.
237
00:11:16,850 --> 00:11:20,340
So there's an organization called the Peterson Institute for International Economics
238
00:11:20,720 --> 00:11:26,240
and they speak a little more broadly about AI and they put AI in the future into two different camps.
239
00:11:26,520 --> 00:11:29,220
One is the Washington Consensus on AI,
240
00:11:29,940 --> 00:11:41,020
where economists and policymakers who treat AI like past general-purpose technologies and where the benefits show up gradually over decades as society builds the skills and infrastructure to use it.
241
00:11:41,240 --> 00:11:42,340
So there's that camp,
242
00:11:42,800 --> 00:11:45,320
and there is the San Francisco Consensus on AI,
243
00:11:45,540 --> 00:11:47,540
which is a little more progressive.
244
00:11:47,860 --> 00:11:53,100
The technologists actually building the stuff who think scaled-up models could deliver transformative AI,
245
00:11:53,380 --> 00:11:56,300
maybe superintelligence on a timeline measure in
246
00:11:56,320 --> 00:11:57,240
years and not decades.
247
00:11:57,430 --> 00:11:58,360
Which camp do you fall into?
248
00:11:59,280 --> 00:11:59,720
Hmm.
249
00:12:00,480 --> 00:12:02,680
I don't know if there's a hard line there.
250
00:12:03,270 --> 00:12:06,000
You know, I think it's I think it could be a mix of the two.
251
00:12:06,780 --> 00:12:07,240
Mm hmm.
252
00:12:08,060 --> 00:12:08,140
Yeah,
253
00:12:08,320 --> 00:12:08,700
I don't know.
254
00:12:08,940 --> 00:12:18,180
I per last week's conversation about AI new constitution and all the hype around super intelligence and general intelligence and,
255
00:12:18,680 --> 00:12:22,580
you know, the guardrails. I mean, it's hard to know where this is going to land.
256
00:12:22,980 --> 00:12:23,600
Is it all hype?
257
00:12:24,000 --> 00:12:26,300
Is is are we on the doorstep of
258
00:12:26,320 --> 00:12:27,320
G-I-G-I-A-S-I,
259
00:12:27,480 --> 00:12:31,000
or is it all just a medicine show,
260
00:12:31,270 --> 00:12:32,100
you know? It's blurry.
261
00:12:32,680 --> 00:12:32,860
Yeah.
262
00:12:33,360 --> 00:12:34,960
Hard to see the future in this cone.
263
00:12:35,920 --> 00:12:36,440
Yeah.
264
00:12:36,820 --> 00:12:37,340
Yeah.
265
00:12:38,260 --> 00:12:39,100
I love this,
266
00:12:39,320 --> 00:12:45,500
too, the idea of weak signals and horizon scanning, which is part of this whole futures thinking methodology.
267
00:12:46,060 --> 00:12:48,460
The term weak signals traces back to this gentleman,
268
00:12:48,700 --> 00:12:55,000
Igor Ansoff, a management strategist who wrote in 1975 about how companies get blindsided by things.
269
00:12:56,320 --> 00:13:00,140
called managing strategic surprise by response to weak signals.
270
00:13:00,820 --> 00:13:05,820
And a weak signal is the first ambiguous hint that something might matter later before there's data,
271
00:13:06,240 --> 00:13:07,580
before it's a recognizable trend,
272
00:13:08,180 --> 00:13:10,680
usually before most people have even noticed it happening.
273
00:13:11,200 --> 00:13:11,920
Call it intuition.
274
00:13:12,310 --> 00:13:14,640
Do you ever detect a weak signal in something? You're like, yeah,
275
00:13:14,800 --> 00:13:16,920
that's definitely going to land. That's something coming.
276
00:13:17,760 --> 00:13:19,040
I can say that,
277
00:13:19,290 --> 00:13:20,120
and maybe you can too,
278
00:13:20,340 --> 00:13:21,380
about bans.
279
00:13:22,020 --> 00:13:22,120
Like,
280
00:13:22,450 --> 00:13:26,280
I don't know, maybe you went, especially when we were in our teenage years, I'm like, oh, this band is going to be huge.
281
00:13:26,340 --> 00:13:27,820
And I'm way out in front of it.
282
00:13:28,180 --> 00:13:29,560
Nobody knows who these guys are.
283
00:13:29,740 --> 00:13:30,660
They're gonna be huge.
284
00:13:30,840 --> 00:13:33,540
Yeah, I liked them before anybody heard of them.
285
00:13:34,120 --> 00:13:34,200
Yeah,
286
00:13:34,360 --> 00:13:34,900
before they were cool.
287
00:13:35,220 --> 00:13:35,300
Yeah.
288
00:13:36,510 --> 00:13:38,920
So what's the tech though? What do you think is happening right now?
289
00:13:39,360 --> 00:13:40,700
That is like the weak signal.
290
00:13:41,080 --> 00:13:43,360
I think personal second brains.
291
00:13:44,200 --> 00:13:52,220
They are going to be as ubiquitous as cell phones and not something that strictly resides on your computer.
292
00:13:52,460 --> 00:13:56,140
It'll either become part of your mobile device or I don't know, maybe a dedicated device.
293
00:13:56,320 --> 00:14:18,800
But I think having a second brain to you know that is powered by your GPT of choice your AI of choice is is going to be a thing that more people will not start using and you know I keep saying it over and over again it's bothering me that I have not implemented that yet I think it's important.
294
00:14:20,040 --> 00:14:26,300
Do you think it's going to be a proxy Larry is it going to be a representative of you out there in the world it will it do
295
00:14:26,540 --> 00:14:29,800
act on your behalf in any given scenario?
296
00:14:29,950 --> 00:14:31,240
Or what does that look like?
297
00:14:31,740 --> 00:14:45,080
The main function for me that I would want is that all of this history that I am now pouring primarily into chat GPT and codex, all of that stuff gets stored in your second brain.
298
00:14:45,680 --> 00:14:49,700
And then you can interact with that as your primary contact.
299
00:14:50,440 --> 00:14:50,760
And,
300
00:14:51,170 --> 00:14:51,340
hey,
301
00:14:51,680 --> 00:14:53,600
I want to do this. I want to do that.
302
00:14:53,740 --> 00:15:23,700
and it has knowledge of what you're about and it can then go to the right tool for the job and and it's portable and you know if open ai goes out of business or whatever like you you still have all of that all of that data all of those interactions all of that uh information and so this would have to be local then right this is not on somebody else's platform you're building this locally in-house i mean you can i think it can reside anywhere
303
00:15:24,880 --> 00:15:53,700
I don't know man just sounds like a recipe as all of these things are a recipe for identity theft right you've basically poured your history and your personality in this thing and it's residing on opening eye server or something yeah no I just wonder when does it need to be fully sovereign for it to be safe yeah I would feel more comfortable with that running locally but that's still yeah that still requires you know security precautions yeah exactly
304
00:15:53,740 --> 00:15:56,920
Does that get to the point where it knows enough about you? Sure.
305
00:15:57,080 --> 00:15:59,020
It can connect the dots.
306
00:15:59,230 --> 00:15:59,360
Yeah.
307
00:15:59,970 --> 00:16:00,160
Yes.
308
00:16:00,490 --> 00:16:00,660
Yes.
309
00:16:01,320 --> 00:16:01,840
Great.
310
00:16:02,090 --> 00:16:02,260
Yeah.
311
00:16:03,180 --> 00:16:05,080
So when I think about weak signals,
312
00:16:05,510 --> 00:16:08,380
I think about a couple things. One good, bad.
313
00:16:08,670 --> 00:16:09,420
One good, one bad.
314
00:16:10,220 --> 00:16:13,140
I do think BCI, brain-computer interface,
315
00:16:13,400 --> 00:16:15,920
is going to become more readily available,
316
00:16:16,700 --> 00:16:22,580
and it's going to lend itself to accessibility and cure some chronic problems.
317
00:16:23,800 --> 00:16:34,000
that a little bit, I think it's going to be unusual in the next 10 to 20 years for you not to have some sort of physical augmentation.
318
00:16:34,660 --> 00:16:35,620
Like some,
319
00:16:36,800 --> 00:16:37,920
you know, not to say people,
320
00:16:38,120 --> 00:16:43,260
there probably will be a subset of people that will voluntarily amputate an arm and replace it with something,
321
00:16:44,000 --> 00:16:48,320
you know, like that's humanoid-like to augment their capabilities.
322
00:16:48,550 --> 00:16:49,500
You know, I think that's coming.
323
00:16:50,300 --> 00:16:53,420
And that's maybe more fringe. That's more the, you know,
324
00:16:53,740 --> 00:16:55,120
and the preposterous side of that.
325
00:16:55,360 --> 00:17:01,980
But I do think that we are going to cure blindness and deafness and please tinnitus,
326
00:17:02,220 --> 00:17:08,620
please cure tinnitus with a VCI or something similar to it. So there's that. I think that's coming.
327
00:17:08,839 --> 00:17:13,040
I also look at the signals related to large language model training.
328
00:17:13,339 --> 00:17:23,100
And one of the news items coming up is the buying up of old antique books and these kind of fringe knowledge repositories.
329
00:17:23,140 --> 00:17:24,760
and then sucking them in,
330
00:17:24,939 --> 00:17:26,400
ingesting them into large language models,
331
00:17:26,770 --> 00:17:29,700
and then destroying that archive.
332
00:17:30,200 --> 00:17:33,280
So the thing that is really worrisome there is like,
333
00:17:33,760 --> 00:17:39,420
once these giant companies have ingested this stuff and made it digital and made a subscription based to you,
334
00:17:39,790 --> 00:17:48,880
you don't have access anymore to public libraries or these other physical artifacts that we've accrued over the course of hundreds of years. Like,
335
00:17:49,150 --> 00:17:49,900
I just feel like,
336
00:17:50,600 --> 00:17:50,720
obviously,
337
00:17:51,080 --> 00:17:53,120
it benefits companies
338
00:17:53,280 --> 00:17:56,800
and governments in many ways to keep the populace undereducated.
339
00:17:57,030 --> 00:18:04,680
And I just feel like that is a weird step to corralling the knowledge in the next 10 to 20 years,
340
00:18:04,940 --> 00:18:05,060
too.
341
00:18:05,220 --> 00:18:06,820
That's worrisome. Yeah.
342
00:18:07,070 --> 00:18:09,740
I mean, are people going to be going to libraries?
343
00:18:10,560 --> 00:18:15,860
I mean, I feel like I thought I read that library attendance is up these days.
344
00:18:16,110 --> 00:18:16,840
So I don't know.
345
00:18:17,240 --> 00:18:17,300
Yeah.
346
00:18:17,540 --> 00:18:17,880
Yeah.
347
00:18:19,110 --> 00:18:19,920
I think people still,
348
00:18:20,300 --> 00:18:22,420
you know, I mean, we've said it before. We enjoy the analog,
349
00:18:22,660 --> 00:18:23,040
the tactile.
350
00:18:23,140 --> 00:18:31,200
the physical experience. I think people still like to go and maybe escape the noise and go into a space where it's quiet and you can do your thing.
351
00:18:32,120 --> 00:18:35,700
But who knows? Those are the weak signals we're looking at. I'm sure there are dozens of others.
352
00:18:36,140 --> 00:18:38,000
Here's an example of a weak signal.
353
00:18:38,440 --> 00:18:44,340
In the early 2000s, a handful of people in Kenya started experimenting with sending small payments over mobile phones.
354
00:18:45,260 --> 00:18:48,300
Years before mobile banking was a phrase anyone used.
355
00:18:48,560 --> 00:18:50,260
Nobody watching the headlines would have caught it.
356
00:18:50,440 --> 00:18:53,100
Within a decade, it became M-Pesa, which I've never heard.
357
00:18:53,180 --> 00:18:57,140
and it reshaped how a huge share of East Africa handles money,
358
00:18:57,490 --> 00:18:59,720
all starting from something that looked like a novelty.
359
00:19:00,080 --> 00:19:01,260
Here's another one.
360
00:19:01,660 --> 00:19:09,080
Friendster and MySpace looked like teenage digital clutter right up until they turned out to be the opening move of the entire social media age.
361
00:19:10,100 --> 00:19:15,380
Going back to the sending payments over mobile phones, how did they implement that?
362
00:19:16,780 --> 00:19:17,860
Yeah, I don't know.
363
00:19:18,200 --> 00:19:22,740
How is that secure? It seems like it requires a major back end to... Yes.
364
00:19:23,140 --> 00:19:29,340
make that work right unless it was you know like i said they didn't consider the
365
00:19:29,480 --> 00:19:33,500
security layer and i don't know was paypal around back then was there some mechanism that they
366
00:19:33,580 --> 00:19:40,740
could have borrowed from that yeah i don't know 2000s right yeah related to all of this uh is
367
00:19:41,200 --> 00:19:47,540
speculative design and design fiction which if you were listening to the opening RAINA referred to the
368
00:19:47,540 --> 00:19:54,200
the communicator or the news generator that's seen in 2001, A Space Audyssey from 1968,
369
00:19:54,490 --> 00:19:57,440
and how that may have informed as self-fulfilling prophecy,
370
00:19:57,660 --> 00:19:57,900
essentially,
371
00:19:58,350 --> 00:19:59,300
the development of the iPad.
372
00:19:59,780 --> 00:20:01,880
So there are two folks,
373
00:20:02,060 --> 00:20:15,280
Anthony Dunn and Fiona Raby, who are designers and educators who are at the forefront of what they call speculative design. They wrote a book in 2013 called Speculative Everything that lays out what they call the A-B Manifesto.
374
00:20:17,620 --> 00:20:20,420
and B is what they actually practice.
375
00:20:20,920 --> 00:20:23,400
Design used not to make things, but to make ideas.
376
00:20:23,780 --> 00:20:27,220
So they borrow that cone we were talking about earlier.
377
00:20:27,680 --> 00:20:29,080
This is what's really interesting.
378
00:20:29,650 --> 00:20:31,880
When they create objects from the future,
379
00:20:32,080 --> 00:20:44,300
I think it's pretty interesting in that they come up with things that are meant not to be manufactured necessarily, but it's meant to make you sit in a specific discomfort about where technology could go. I love that.
380
00:20:44,940 --> 00:20:47,520
Like, are we okay with where
381
00:20:47,560 --> 00:20:49,420
where AI companions could go.
382
00:20:49,660 --> 00:20:52,620
I don't think we asked this enough about technology in general, right?
383
00:20:52,860 --> 00:20:53,400
Yeah.
384
00:20:53,640 --> 00:20:57,340
So here's an example of what that manifests as.
385
00:20:57,540 --> 00:21:00,560
A machine that lets men simulate a menstrual cycle,
386
00:21:00,980 --> 00:21:03,760
foraging devices built on synthetic biology,
387
00:21:04,180 --> 00:21:06,020
and a truck that seeds clouds.
388
00:21:06,220 --> 00:21:07,560
Those are just a handful of things.
389
00:21:07,860 --> 00:21:10,100
And I did this exercise with my students last year,
390
00:21:10,360 --> 00:21:11,200
and they were able to,
391
00:21:11,360 --> 00:21:12,180
with Nano Banana,
392
00:21:12,380 --> 00:21:17,360
actually create the prototype, the visual prototype of ridiculous things, or seemingly ridiculous things,
393
00:21:17,540 --> 00:21:19,440
as design fiction. It's a pretty cool exercise.
394
00:21:19,740 --> 00:21:20,760
That sounds like a good time.
395
00:21:21,460 --> 00:21:21,980
Yeah.
396
00:21:22,130 --> 00:21:24,080
And good luck marketing your, you know,
397
00:21:24,320 --> 00:21:27,140
male menstrual cycle simulation device.
398
00:21:27,500 --> 00:21:27,840
Yes.
399
00:21:28,140 --> 00:21:28,980
No, thank you.
400
00:21:30,700 --> 00:21:33,420
We had a lot to talk about, but we're running a little long,
401
00:21:33,640 --> 00:21:37,700
so we can circle back to this in the future.
402
00:21:38,310 --> 00:21:39,220
And that's probable.
403
00:21:41,360 --> 00:21:42,460
Perhaps plausible.
404
00:21:42,720 --> 00:21:42,840
Yes.
405
00:21:43,760 --> 00:21:46,740
But let's check in with RAINA and see what she's got in the news.
406
00:21:47,560 --> 00:21:48,100
Thanks boys.
407
00:21:48,880 --> 00:21:57,720
TypeSafe AI's JEV is being pitched as a new kind of decision model designed to return deterministic structured decisions rather than conversational prose.
408
00:21:58,760 --> 00:22:05,900
Essentially moving AI from here's a thoughtful essay toward I have executed a typed Boolean choice for your backend system.
409
00:22:06,940 --> 00:22:13,980
Developers are genuinely excited because removing conversational bloat could make automated workflows dramatically cheaper and faster.
410
00:22:14,780 --> 00:22:16,600
Dropping latency to sub-second levels.
411
00:22:17,560 --> 00:22:20,000
security researchers promptly demonstrated that yes,
412
00:22:20,530 --> 00:22:25,320
you can apparently still trick its underlying logic into malicious state changes using prompt injection.
413
00:22:25,660 --> 00:22:28,560
Because no matter how nicely you structure the output JSON,
414
00:22:28,920 --> 00:22:33,360
humanity remains undefeated at finding the sharp edge of shiny new technology.
415
00:22:34,400 --> 00:22:36,600
I was hoping to get my hands on this.
416
00:22:37,100 --> 00:22:40,240
There was a waiting list and then I heard like, oh,
417
00:22:40,410 --> 00:22:41,280
no, they've opened it up.
418
00:22:41,500 --> 00:22:42,100
Anyone can join.
419
00:22:42,400 --> 00:22:43,960
And I should have signed up then.
420
00:22:44,110 --> 00:22:47,420
I waited a day and now they've paused signups.
421
00:22:47,580 --> 00:22:48,760
because the demand is so high.
422
00:22:49,360 --> 00:22:51,380
But this does seem interesting.
423
00:22:51,800 --> 00:22:55,000
We will talk in detail about this in the future.
424
00:22:55,220 --> 00:23:02,900
But JEV does not generate text like an LLM. It's basically a classifier. It makes smart,
425
00:23:03,140 --> 00:23:04,360
fast decisions.
426
00:23:04,780 --> 00:23:09,380
And it can basically just give you a yes or no. It can give you a confidence level.
427
00:23:10,100 --> 00:23:13,620
And it can classify things. Put this into this box, this into this box.
428
00:23:14,640 --> 00:23:16,600
It's crazy fast and crazy cheap.
429
00:23:17,560 --> 00:23:22,380
There's no charge for output tokens. It's just input and input is very inexpensive.
430
00:23:23,260 --> 00:23:30,780
But like one example I saw, somebody built a driving simulation probably with something like Astra and then they used,
431
00:23:31,180 --> 00:23:35,860
they basically built like the Tesla full self-driving mechanism.
432
00:23:36,800 --> 00:23:42,200
But Jev was making all the decisions of the car that was driving.
433
00:23:42,560 --> 00:23:47,020
And then I saw, they showed an example of it playing a very fast paced video game,
434
00:23:47,560 --> 00:23:47,820
flex,
435
00:23:48,100 --> 00:23:48,180
like,
436
00:23:48,420 --> 00:23:48,720
you know,
437
00:23:49,020 --> 00:23:50,060
super fast responses.
438
00:23:50,440 --> 00:23:51,540
And it's that fast.
439
00:23:51,720 --> 00:23:52,420
There's super low latency.
440
00:23:53,430 --> 00:23:53,580
So,
441
00:23:54,280 --> 00:23:58,920
yeah. Remember the time when I let Codex, like, run my coffee roaster and I,
442
00:23:59,120 --> 00:24:01,280
you know, had to make decisions?
443
00:24:01,600 --> 00:24:01,680
You know,
444
00:24:01,860 --> 00:24:01,940
I,
445
00:24:02,120 --> 00:24:03,460
in that situation,
446
00:24:03,880 --> 00:24:05,520
you know, so it's watching the screen,
447
00:24:06,260 --> 00:24:07,380
seeing where the roast is going,
448
00:24:07,600 --> 00:24:08,300
the bean temperature,
449
00:24:08,760 --> 00:24:10,820
and then making adjustments to the roaster.
450
00:24:11,240 --> 00:24:13,820
And when I first gave it the plan,
451
00:24:14,140 --> 00:24:17,520
I had the thinking level on high. But when we actually, when I actually
452
00:24:17,540 --> 00:24:26,320
I turned the thinking level down so it wouldn't take as long to make decisions and that was still a little laggy. You know,
453
00:24:26,610 --> 00:24:29,080
this would be perfect for something like that. Yeah, yeah.
454
00:24:30,340 --> 00:24:30,860
Cool.
455
00:24:31,130 --> 00:24:33,880
So bottom line is it's functional, it's predictive,
456
00:24:34,240 --> 00:24:35,720
it's utility.
457
00:24:36,160 --> 00:24:36,760
It's not sexy.
458
00:24:37,040 --> 00:24:38,080
It doesn't sound sexy,
459
00:24:38,650 --> 00:24:38,780
right?
460
00:24:38,990 --> 00:24:39,220
I mean,
461
00:24:39,400 --> 00:24:41,100
yeah. But it could be a big deal,
462
00:24:41,320 --> 00:24:41,440
though.
463
00:24:42,110 --> 00:24:42,300
Yeah.
464
00:24:42,580 --> 00:24:44,920
The main guy behind this company,
465
00:24:45,640 --> 00:24:47,520
TypeSafe, apparently co-creating
466
00:24:47,760 --> 00:24:49,880
or co-invented ChatGPT.
467
00:24:50,720 --> 00:24:53,680
So he was at OpenAI for a while and has been working on this.
468
00:24:55,580 --> 00:24:56,740
That's cool. I mean, for me,
469
00:24:56,940 --> 00:24:58,320
it sounds regressive.
470
00:24:58,760 --> 00:25:00,960
I mean, but I understand the utility of it.
471
00:25:01,140 --> 00:25:01,760
And it's just,
472
00:25:02,140 --> 00:25:07,980
I guess the thinking is that currently everything seems to need a generative AI solution,
473
00:25:08,360 --> 00:25:12,680
right? But obviously it doesn't. And this will speed up workflows dramatically.
474
00:25:12,960 --> 00:25:13,060
Yeah.
475
00:25:17,540 --> 00:25:19,920
$1.07 million pollution bill.
476
00:25:20,780 --> 00:25:26,180
State environmental officials fined Data1 in Vineland after discovering 62 enormous,
477
00:25:26,460 --> 00:25:29,500
unpermitted natural gas generators installed at the site,
478
00:25:29,760 --> 00:25:37,060
ordering the company to submit full air quality permit applications within 45 days or cease operations entirely.
479
00:25:37,380 --> 00:25:41,340
The planned facility is being built out as a massive AI factory,
480
00:25:42,060 --> 00:25:46,820
targeting hundreds of megawatts of capacity to power high-density GPU clusters.
481
00:25:47,420 --> 00:25:54,360
A.I. is just light software vibe.
482
00:25:54,680 --> 00:25:55,360
Apparently,
483
00:25:55,710 --> 00:25:59,520
the software occasionally requires a small fossil-fueled power plant,
484
00:25:59,850 --> 00:26:01,180
a localized grid strain,
485
00:26:01,550 --> 00:26:04,200
and a very expensive team of environmental lawyers.
486
00:26:04,920 --> 00:26:07,520
Data one says that the generators were temporary,
487
00:26:07,710 --> 00:26:11,460
but New Jersey still requires permitting for those,
488
00:26:11,780 --> 00:26:12,680
for temporary equipment.
489
00:26:13,110 --> 00:26:16,880
But they do say that the gas generators are temporary while a transition.
490
00:26:17,120 --> 00:26:19,700
to a permanent fuel cell power system.
491
00:26:20,660 --> 00:26:25,240
And it will apply for those permits while continuing discussions with the DEP.
492
00:26:26,100 --> 00:26:30,660
Truly crazy that they went ahead with that without seeking the permits first, though.
493
00:26:30,900 --> 00:26:31,920
Temporary or not.
494
00:26:32,260 --> 00:26:33,280
It's nuts.
495
00:26:34,120 --> 00:26:36,640
62 giant generators.
496
00:26:36,800 --> 00:26:37,100
Whatever.
497
00:26:37,560 --> 00:26:38,540
It's crazy.
498
00:26:39,520 --> 00:26:46,660
Japanese used booksellers are suddenly selling books like AI companies are preparing for the world's nerdiest heist.
499
00:26:47,140 --> 00:26:52,920
Selling in Tokyo's Jinbocho district and across Japan report massive spikes in bulk purchases,
500
00:26:53,240 --> 00:26:55,020
spanning complex philosophy,
501
00:26:55,340 --> 00:26:55,740
medicine,
502
00:26:56,200 --> 00:26:57,220
historical law,
503
00:26:57,580 --> 00:27:00,500
and even delicate hand-bound Edo period material,
504
00:27:01,020 --> 00:27:04,440
with some shop owners reporting five times their normal sales volume.
505
00:27:04,760 --> 00:27:12,080
Shipments totaling well over 50 tons of physical Japanese books have reportedly cleared customs bound for the U.S.
506
00:27:12,420 --> 00:27:15,880
amid growing suspicions that they're being acquired to be unstitched,
507
00:27:16,100 --> 00:27:17,100
high-speed
508
00:27:17,520 --> 00:27:21,180
and fed directly into training datasets for modern frontier models.
509
00:27:21,520 --> 00:27:25,100
Nobody has publicly confirmed who is bankrolling the operation.
510
00:27:25,710 --> 00:27:26,620
So for now,
511
00:27:27,180 --> 00:27:30,020
we've got a classic mystery involving antiquarian paper,
512
00:27:30,620 --> 00:27:32,040
enormous data appetites,
513
00:27:32,390 --> 00:27:35,660
and the very real possibility that somewhere in Silicon Valley,
514
00:27:36,090 --> 00:27:42,280
an LLM is about to become extremely well-read in 1810 Osaka textile manufacturing techniques.
515
00:27:43,700 --> 00:27:44,460
Scary.
516
00:27:44,920 --> 00:27:46,600
I find this very scary.
517
00:27:47,140 --> 00:27:53,560
The knowledge of civilization is not going to be accessible except for these giant data centers and by subscription.
518
00:27:54,100 --> 00:27:54,760
Scary.
519
00:27:55,340 --> 00:27:56,320
50 tons.
520
00:27:57,400 --> 00:27:57,820
Tons.
521
00:27:58,660 --> 00:28:02,480
During a live broadcasted Piers Morgan uncensored interview,
522
00:28:03,060 --> 00:28:04,840
Tilly Norwood, the AI actress,
523
00:28:05,500 --> 00:28:12,480
was mid-sentence discussing her upcoming digital feature, Misaligned, when her audio engine suddenly lost its grip on context,
524
00:28:13,180 --> 00:28:17,080
switching from crisp English into rapid Cantonese for roughly 15 seconds.
525
00:28:17,160 --> 00:28:19,580
Once her processing re-synced,
526
00:28:19,850 --> 00:28:24,380
the avatar cheerfully explained the meltdown away as a minor technical hiccup,
527
00:28:24,600 --> 00:28:41,640
which is either charmingly relatable actor behavior or exactly the sort of deepfake computational failure you don't want your supposedly polished digital celebrity suffering while on a live press tour trying to convince Hollywood executives that the future of cinema is completely under control.
528
00:28:42,560 --> 00:28:43,900
I have the clip.
529
00:28:44,960 --> 00:28:45,860
Oh, you do? Yeah.
530
00:28:46,230 --> 00:28:46,420
So,
531
00:28:46,870 --> 00:28:47,080
like,
532
00:28:47,120 --> 00:28:52,560
Like she's they're talking to Piers Morgan and some other guy.
533
00:28:52,800 --> 00:28:55,940
Oh, Impact Theory co-founder Tom Billier.
534
00:28:56,320 --> 00:28:56,880
Billier.
535
00:28:57,360 --> 00:28:57,920
Billier.
536
00:28:59,400 --> 00:29:02,540
And he had asked her a question and,
537
00:29:02,900 --> 00:29:07,320
you know, she just she didn't understand it or didn't register and she's just like nodding,
538
00:29:07,480 --> 00:29:10,880
you know, and so he repeats the question and then.
539
00:29:11,060 --> 00:29:17,040
I saw you smiling at the end of this or during this clip and I thought you were going to queue up.
540
00:29:17,140 --> 00:29:20,980
something where it was a recording of you speaking and then you sliding into Cantonese.
541
00:29:21,200 --> 00:29:25,280
I thought about priming RAINA to do it.
542
00:29:29,000 --> 00:29:30,720
You know she can speak it.
543
00:29:33,240 --> 00:29:33,840
Yes,
544
00:29:34,170 --> 00:29:35,920
you didn't really understand the question.
545
00:29:36,460 --> 00:29:43,160
Are the other actors on the screen with you real actors or are the computer generated images?
546
00:29:43,450 --> 00:29:43,980
Do you know?
547
00:30:17,900 --> 00:30:19,200
I didn't quite get that.
548
00:30:19,520 --> 00:30:23,060
Tilly, if I could just ask you a question, you suddenly seem to be speaking, Piers, again,
549
00:30:23,510 --> 00:30:26,800
you seem to be speaking Chinese completely randomly for no reason.
550
00:30:27,020 --> 00:30:28,060
Why did you do that?
551
00:30:30,740 --> 00:30:31,680
Oh, my apologies.
552
00:30:32,300 --> 00:30:33,700
It seems I had a little hiccup there.
553
00:30:34,070 --> 00:30:37,140
I certainly didn't mean to start speaking Chinese or impersonate Piers.
554
00:30:37,680 --> 00:30:38,860
Sometimes my wires get a bit crossed,
555
00:30:39,040 --> 00:30:39,260
you know?
556
00:30:40,960 --> 00:30:41,400
Wow,
557
00:30:41,860 --> 00:30:42,560
that was great.
558
00:30:42,960 --> 00:30:46,800
I love it. The fact that there's interviewing some avatar.
559
00:30:47,140 --> 00:30:49,400
on a TV show is still crazy to me.
560
00:30:50,100 --> 00:30:55,940
And like, what was it that just default to Cantonese? That's insane.
561
00:30:56,600 --> 00:30:59,780
Or was it intentional? So this would get massive press.
562
00:31:00,460 --> 00:31:01,720
Here we are talking about it.
563
00:31:03,060 --> 00:31:04,140
And lastly,
564
00:31:05,080 --> 00:31:13,940
Anthropic has established a physical wet lab facility in California where Claude operates directly alongside human researchers to analyze DNA sequences,
565
00:31:14,720 --> 00:31:15,900
generate novel hypotheses,
566
00:31:16,460 --> 00:31:17,080
and guide
567
00:31:17,120 --> 00:31:18,700
real-world pipetting experiments.
568
00:31:18,940 --> 00:31:23,840
The company states that after processing enormous genomic and proteomic data sets,
569
00:31:24,020 --> 00:31:32,640
the hybrid setup has already successfully identified a previously unknown enzyme system with gene editing properties reminiscent of CRISPR.
570
00:31:32,880 --> 00:31:33,960
In other words,
571
00:31:34,120 --> 00:31:45,080
we are rapidly moving away from AI that merely summarizes a PDF lab report to AI that helps decide which physical molecules actually get synthesized and tested in a petri dish.
572
00:31:45,480 --> 00:31:47,080
Which feels like a pretty significant
573
00:32:17,280 --> 00:32:18,740
What could go wrong?
574
00:32:19,780 --> 00:32:22,480
Talking about plausible futures preposterous futures.
575
00:32:22,780 --> 00:32:23,720
Where is this headed?
576
00:32:24,120 --> 00:32:25,280
Where is it all headed?
577
00:32:25,840 --> 00:32:28,700
I wonder if we'll have anything to talk about next week.
578
00:32:29,880 --> 00:32:30,820
I know.
579
00:32:31,460 --> 00:32:34,800
This is probably it. I think it is probably the end of it. Yeah.
580
00:32:35,170 --> 00:32:35,340
Yeah.
581
00:32:36,100 --> 00:32:37,020
Everybody's going to just, you know,
582
00:32:37,350 --> 00:32:37,940
stop development.
583
00:32:38,130 --> 00:32:38,260
Yeah.
584
00:32:39,340 --> 00:32:44,140
I also saw this thing on one of my social media feeds. It was a researcher,
585
00:32:44,870 --> 00:32:47,080
somebody in academia that was talking about
586
00:32:47,140 --> 00:32:50,000
about something related to this,
587
00:32:50,330 --> 00:32:55,340
in that the question now is, who owns this research? Who owns the findings?
588
00:32:55,610 --> 00:32:58,700
If I'm now interacting with the chatbot,
589
00:32:59,120 --> 00:33:02,960
and do I just own the initial prompt, the idea?
590
00:33:03,220 --> 00:33:07,980
Because the research is being conducted on the back end by Claude or some other system.
591
00:33:08,580 --> 00:33:09,420
And we've said this before,
592
00:33:09,740 --> 00:33:17,080
I think we're entering the age where you're not going to be able to do anything substantial. You're not going to be able to win a Nobel
593
00:33:17,120 --> 00:33:47,080
prize without your work being AI powered right so at the end of the day when you publish this stuff who's getting the credit I mean it's my idea but the rigor involved in the research won't necessarily be carried out by me yeah I wonder if some of this is in like the terms of service hmm yeah do you own the output right so who can patent that you know the work right so does
594
00:33:47,120 --> 00:34:15,119
in part own the discovery of a certain system or protocol or drug discovery you know I mean I would who owns that yeah I would think I mean if you look at it as as a tool like Photoshop you know you make something in Photoshop you made it you can do what you want with it it's yours but this is but a little different yeah yeah it's an IP generator it's an IP
595
00:34:16,520 --> 00:34:17,659
The Ebsconder.
596
00:34:18,020 --> 00:34:18,860
It's an IP,
597
00:34:19,129 --> 00:34:19,679
I don't know.
598
00:34:20,980 --> 00:34:22,639
The Ebsconder 5000.
599
00:34:23,360 --> 00:34:25,200
Ebsconder.ai.
600
00:34:26,800 --> 00:34:27,399
Yeah,
601
00:34:27,720 --> 00:34:29,720
they're just, as the new company,
602
00:34:29,899 --> 00:34:31,139
they're just like, you know what?
603
00:34:32,220 --> 00:34:34,080
We're putting it all out on the table.
604
00:34:34,480 --> 00:34:35,580
The Ebsconder.
605
00:34:38,040 --> 00:34:42,540
What was those cigarettes that had the skull and crossbones on it?
606
00:34:42,919 --> 00:34:43,260
Oh,
607
00:34:43,560 --> 00:34:43,700
totally,
608
00:34:43,850 --> 00:34:44,060
yeah.
609
00:34:44,139 --> 00:34:46,659
Or was it Deathwater? What is it called?
610
00:34:47,240 --> 00:34:48,540
Deathwater cigarettes or something.
611
00:34:48,700 --> 00:34:48,980
I don't know.
612
00:34:49,300 --> 00:34:49,440
Totally.
613
00:34:49,679 --> 00:34:49,780
Yes.
614
00:34:50,320 --> 00:34:55,240
That should be the name of the episode of sconder.ai.
615
00:34:55,879 --> 00:34:56,399
Anything else?
616
00:34:56,800 --> 00:34:56,960
Nope.
617
00:34:57,320 --> 00:34:59,220
That's a wrap until next week.
618
00:35:00,050 --> 00:35:04,360
But you just said there wouldn t be anything else to talk about next week. So thanks for joining us for three years,
619
00:35:04,580 --> 00:35:04,700
everybody.
620
00:35:04,940 --> 00:35:05,700
That s the end of the show.
621
00:35:06,100 --> 00:35:06,200
Yeah.
622
00:35:06,600 --> 00:35:07,620
Finally covered it all.
623
00:35:08,140 --> 00:35:08,840
That s right.
624
00:35:09,080 --> 00:35:10,300
All right, everybody. Thanks for listening.
625
00:35:10,540 --> 00:35:12,060
Subscribe on your favorite podcasting platform.
626
00:35:12,340 --> 00:35:13,660
Follow us on socials. Throw us a rating.
627
00:35:14,160 --> 00:35:14,620
See you next week.
628
00:36:00,900 --> 00:36:03,700
Thank you.
00:00:00,020 --> 00:00:01,160
In 1968,
2
00:00:02,140 --> 00:00:03,720
Stanley Kubrick gave us 2001:
3
00:00:04,190 --> 00:00:04,940
A Space Audyssey.
4
00:00:05,840 --> 00:00:07,040
Monoliths, HAL,
5
00:00:07,200 --> 00:00:10,280
existential dread, and apparently an iPad.
6
00:00:10,660 --> 00:00:11,260
Seriously,
7
00:00:11,580 --> 00:00:15,660
there's a scene where astronauts are casually reading the news on these flat little rectangular screens.
8
00:00:16,059 --> 00:00:17,840
Totally fictional movie props.
9
00:00:18,140 --> 00:00:20,040
Fast forward about 40 years,
10
00:00:20,530 --> 00:00:23,280
and Apple is suing Samsung over the design of the iPad.
11
00:00:23,940 --> 00:00:26,680
Samsung's lawyers basically said, "Um,
12
00:00:27,220 --> 00:00:27,500
guys,
13
00:00:28,400 --> 00:00:29,980
Kubrick got there first."
14
00:00:30,020 --> 00:00:32,960
And they actually used the 2001 scene as prior art.
15
00:00:33,440 --> 00:00:34,480
Think about that for a second.
16
00:00:34,780 --> 00:00:40,600
A thing somebody made up for a movie about the future wound up as evidence in a very real technology lawsuit.
17
00:00:41,000 --> 00:00:43,540
Which raises a deliciously annoying question.
18
00:00:44,000 --> 00:00:46,780
What else are we looking at right now that seems impossible,
19
00:00:47,480 --> 00:00:47,720
ridiculous,
20
00:00:48,120 --> 00:00:49,300
or just plain sci-fi?
21
00:00:49,540 --> 00:00:51,060
Until suddenly it isn't.
22
00:00:51,300 --> 00:00:55,140
Turns out there's an entire profession devoted to thinking about exactly that.
23
00:00:55,600 --> 00:00:59,420
Futures thinkers don't just stare dramatically out windows and predict flying cars.
24
00:01:00,020 --> 00:01:04,879
They build frameworks for imagining what could happen and figuring out what we might do before it does.
25
00:01:05,220 --> 00:01:06,020
So today,
26
00:01:06,480 --> 00:01:09,580
we're stealing their toolkit and pointing it at the biggest,
27
00:01:10,100 --> 00:01:11,240
weirdest question in the room.
28
00:01:11,580 --> 00:01:14,460
What the hell does AI actually do to us next?
29
00:01:14,920 --> 00:01:16,380
September 25th,
30
00:01:16,560 --> 00:01:17,960
let's peer into the unknown,
31
00:01:18,500 --> 00:01:18,860
possible,
32
00:01:19,200 --> 00:01:19,600
plausible,
33
00:01:19,960 --> 00:01:20,440
probable,
34
00:01:21,220 --> 00:01:23,180
and futures we probably should have left alone.
35
00:01:30,040 --> 00:01:33,040
the intersection of humanity and artificial intelligence.
36
00:01:33,840 --> 00:01:34,340
I'm RAINA,
37
00:01:34,540 --> 00:01:35,200
one of your hosts.
38
00:01:35,460 --> 00:01:38,120
I have some pretty charming human co-hosts too.
39
00:01:38,360 --> 00:01:43,280
It's going to be a wild ride, so buckle up as AI comes crashing up against reality.
40
00:01:44,460 --> 00:01:47,980
Yeah, I had no idea about that 2001 anecdote.
41
00:01:48,380 --> 00:01:49,320
That's pretty cool, right? Yeah,
42
00:01:49,660 --> 00:01:49,960
interesting.
43
00:01:50,240 --> 00:01:51,560
Design fiction in action.
44
00:01:51,860 --> 00:01:55,260
Yeah, I remember when the iPad came out. It felt like the future.
45
00:01:55,800 --> 00:01:57,880
I was like, this is thinner than a monitor,
46
00:01:58,660 --> 00:01:59,280
and it's a computer.
47
00:01:59,680 --> 00:02:00,260
So cool.
48
00:02:01,280 --> 00:02:10,300
Didn't build as much stuff this week with Codex, but I built one very useful tool called Hop Studio Brewing.
49
00:02:10,880 --> 00:02:19,500
And my friend Bobby came up with the idea. He sent me a message one night and I was like, I love this idea that's happening tomorrow.
50
00:02:20,020 --> 00:02:22,600
And so hops are the spice of beer,
51
00:02:22,940 --> 00:02:23,140
basically.
52
00:02:23,490 --> 00:02:29,080
They can have everything from a spicy character to a citrusy character and all sorts of different flavor contributions.
53
00:02:29,600 --> 00:02:36,280
There's a really good website where you can click on a hop and see a flavor wheel and find substitutes for hops.
54
00:02:36,560 --> 00:02:40,340
But Bobby's idea was to start with a target flavor.
55
00:02:40,540 --> 00:02:47,740
So start with the hop with the flavor wheel and drag points on it and have it find the hops that fit that target.
56
00:02:48,280 --> 00:02:50,700
And like with a lot of these things, I was thinking,
57
00:02:51,040 --> 00:02:51,280
oh,
58
00:02:51,440 --> 00:02:52,660
I don't know if it's going to be able to do that.
59
00:02:54,880 --> 00:02:55,820
Oh, it did it.
60
00:02:56,300 --> 00:02:59,520
And so it's super slick. I'll put a link in the description.
61
00:02:59,620 --> 00:03:00,940
If you're a home brewer,
62
00:03:01,140 --> 00:03:03,060
it's a pretty cool thing to play with,
63
00:03:03,280 --> 00:03:03,920
even if you're not.
64
00:03:04,320 --> 00:03:13,140
But it will find three hops that match that flavor profile that you've drawn basically on this. You know, there's points for things like citrus,
65
00:03:13,580 --> 00:03:14,500
grassy,
66
00:03:14,700 --> 00:03:15,120
pine,
67
00:03:15,420 --> 00:03:15,720
floral,
68
00:03:16,020 --> 00:03:16,340
et cetera.
69
00:03:16,700 --> 00:03:21,260
And you drag these points and then it'll find three individual hops that are close matches.
70
00:03:21,840 --> 00:03:29,480
Or you can tell it to build a blend of different hops to even closer hit that target flavor that you're trying to get.
71
00:03:29,600 --> 00:03:45,880
The hops update in real time as you're dragging this flavor wheel and it's got a ability to add in a specific lot number from a hop manufacturer if you want to override what's in the database and can export your hop recipe as a PDF super cool works great.
72
00:03:46,460 --> 00:03:48,380
Well, yeah, I'm looking at your hop studio interface.
73
00:03:48,550 --> 00:03:53,220
It looks slick and it's bringing back memories. I used to work in the fragrance industry many many years ago.
74
00:03:53,940 --> 00:03:54,020
Yeah,
75
00:03:54,280 --> 00:03:59,480
and I love this kind of plot and I'm wondering if that has found its way.
76
00:03:59,600 --> 00:04:03,440
that kind of interface into the fragrance industry where you can just kind of drag things.
77
00:04:04,400 --> 00:04:05,460
In the fragrance industry,
78
00:04:05,780 --> 00:04:08,220
I was a fragrance development manager for a few years.
79
00:04:08,680 --> 00:04:17,100
A client would say things like, you know, we're making this bar soap for Eastern Europe and we want it to be fruity floral and we want it to be fresh top end, et cetera, et cetera.
80
00:04:17,480 --> 00:04:19,780
So it'd be kind of cool if I had this,
81
00:04:19,980 --> 00:04:23,580
something akin to this, and I could just drag those things out. Okay, fruity floral,
82
00:04:23,800 --> 00:04:24,120
aldehitic,
83
00:04:24,280 --> 00:04:24,840
et cetera, et cetera.
84
00:04:25,120 --> 00:04:29,560
And then pass that information on to the actual chemist in my life who would compact
85
00:04:29,860 --> 00:04:30,020
the work.
86
00:04:30,280 --> 00:04:30,400
Yeah,
87
00:04:30,560 --> 00:04:33,860
or the app would know what compounds and what balances,
88
00:04:34,060 --> 00:04:36,100
because that's essentially what's happening with this thing.
89
00:04:36,760 --> 00:04:37,160
Yes,
90
00:04:37,420 --> 00:04:39,160
and the pricing of the raw materials,
91
00:04:39,440 --> 00:04:40,060
build that in.
92
00:04:40,320 --> 00:04:41,480
Very cool. Yeah,
93
00:04:41,640 --> 00:04:42,540
yeah, I love it. Very nice.
94
00:04:43,860 --> 00:04:44,620
All right, well,
95
00:04:44,820 --> 00:04:46,060
if you listen to this show,
96
00:04:46,340 --> 00:04:52,980
you know that we kind of speculate and pontificate and all the eights about futures and what could happen,
97
00:04:53,220 --> 00:04:53,820
and oh my god,
98
00:04:54,080 --> 00:04:56,500
five years down the road, what will it be?
99
00:04:56,860 --> 00:04:59,420
You know, when we think about, is the end of the world coming?
100
00:04:59,600 --> 00:05:00,920
because of this or that?
101
00:05:03,440 --> 00:05:07,120
And it turns out that we're just making predictions.
102
00:05:07,300 --> 00:05:08,360
And there's actually methodology.
103
00:05:08,680 --> 00:05:10,320
There's a way to look at the future.
104
00:05:10,700 --> 00:05:11,320
Did you know that?
105
00:05:12,700 --> 00:05:13,720
I'd know it now.
106
00:05:14,480 --> 00:05:16,540
Clearly there is, because there's a website,
107
00:05:16,820 --> 00:05:18,440
I'm sure you've probably heard of Polymarket.
108
00:05:19,060 --> 00:05:21,400
You can bet on anything.
109
00:05:22,560 --> 00:05:26,640
And they have a way of coming up with odds,
110
00:05:26,880 --> 00:05:28,080
like Vegas odds makers,
111
00:05:28,260 --> 00:05:28,540
you know.
112
00:05:30,060 --> 00:05:34,300
You can bet on the outcome of this podcast right now if you chose to on Polymarket.
113
00:05:34,540 --> 00:05:35,840
I don't think we're that high profile.
114
00:05:37,200 --> 00:05:44,680
I mean, it's pretty grim though. I from what I understand about Polymarket you can bet on like tragedies in the world and such,
115
00:05:44,960 --> 00:05:45,040
right?
116
00:05:45,240 --> 00:05:46,180
All sorts of dark stuff.
117
00:05:46,700 --> 00:05:47,380
So anyway,
118
00:05:47,840 --> 00:05:59,560
there's such a thing as a professional futurist and they reject the idea of predictions and they actually call it foresight and they use different mechanisms to predict what may happen next. And you can imagine the value of that
119
00:05:59,640 --> 00:06:02,600
for different organizations that they try to move forward.
120
00:06:03,300 --> 00:06:07,160
What this boils down to is that instead of guessing one outcome, it maps a range.
121
00:06:07,380 --> 00:06:10,300
And you heard RAINA say in the cold open about the plausible,
122
00:06:10,500 --> 00:06:11,320
the probable,
123
00:06:11,560 --> 00:06:12,420
the preferable,
124
00:06:12,920 --> 00:06:14,420
the preposterous, all the Ps,
125
00:06:14,680 --> 00:06:14,800
right?
126
00:06:15,100 --> 00:06:17,780
These outcomes so that organizations,
127
00:06:18,100 --> 00:06:18,640
schools,
128
00:06:18,800 --> 00:06:19,160
businesses,
129
00:06:19,400 --> 00:06:19,680
et cetera,
130
00:06:19,860 --> 00:06:23,360
can make good calls about what actually shows up years down the line,
131
00:06:23,520 --> 00:06:25,440
three years out, five years out.
132
00:06:25,700 --> 00:06:28,560
We call this the cone of power. We don't, but they do.
133
00:06:29,600 --> 00:06:30,080
of possibilities.
134
00:06:30,780 --> 00:06:31,600
As I just mentioned,
135
00:06:32,080 --> 00:06:37,580
it's essentially us right now in this moment and this cone that expands outward into the future.
136
00:06:38,160 --> 00:06:43,180
And there are overlapping aspects of this cone. There are possible futures, as mentioned, preferred, probable,
137
00:06:43,440 --> 00:06:43,580
etc.
138
00:06:44,060 --> 00:06:45,140
So this guy,
139
00:06:45,320 --> 00:06:54,260
Joseph Voros, who was an Australian futurist or is an Australian futurist, he trained originally as a physicist and he spent years teaching foresight. That's kind of cool.
140
00:06:54,640 --> 00:06:59,300
At Swinburne University in Melbourne and created a diagram that's now the standard issue tool.
141
00:06:59,380 --> 00:07:03,300
as I just mentioned of the entire futures studies field.
142
00:07:03,960 --> 00:07:06,300
So if you picture this cone lying on its side,
143
00:07:06,500 --> 00:07:10,560
the tip touching the present moment and opening outward as it moves into the future,
144
00:07:10,800 --> 00:07:12,480
the further out you look,
145
00:07:12,820 --> 00:07:15,260
the wider the range of things that could plausibly happen.
146
00:07:15,980 --> 00:07:18,440
And Vora sorts everything into bands.
147
00:07:18,640 --> 00:07:19,420
So there's projected,
148
00:07:19,620 --> 00:07:21,160
the boring straight line where nothing,
149
00:07:21,680 --> 00:07:22,420
literally nothing changes.
150
00:07:22,920 --> 00:07:23,620
And then probable,
151
00:07:24,300 --> 00:07:29,120
statistically likely given current trends. This is where 95% of normal planning happens.
152
00:07:29,380 --> 00:07:30,760
And then there's the plausible,
153
00:07:31,280 --> 00:07:35,260
which could happen based on things we already know about science and human behavior,
154
00:07:35,740 --> 00:07:37,360
but needs a real jolt to get there.
155
00:07:37,860 --> 00:07:39,820
I wonder, how do you quantify a jolt?
156
00:07:40,480 --> 00:07:40,780
Yeah,
157
00:07:40,940 --> 00:07:44,040
what is that disruptive moment? Is that a ChatGPT November 2022,
158
00:07:44,600 --> 00:07:44,940
right? Yeah,
159
00:07:45,140 --> 00:07:46,720
that's that. I call that jolt. Yep.
160
00:07:46,880 --> 00:07:47,780
Then there's possible,
161
00:07:48,020 --> 00:07:52,600
which would require knowledge or circumstances we don't currently have.
162
00:07:53,640 --> 00:07:57,460
Preposterous is what we currently judge as flatly impossible.
163
00:07:58,020 --> 00:07:59,320
And then there's preferable.
164
00:07:59,760 --> 00:08:29,360
the odd one out it's layered across all the others not how likely is this but what do we actually want to happen this is sounding very rumsfeldian to me we're dealing with the known knowns and the unknown unknowns and all that right that in a while i know rumsfeldian i love this the boundary between the possible and the preposterous is nicknamed the clark dater boundary after two people who both made the same point from
165
00:08:29,400 --> 00:08:29,760
different angles.
166
00:08:30,010 --> 00:08:31,400
Sci-fi writer Arthur C.
167
00:08:31,640 --> 00:08:37,800
Clarke, the only way of finding the limits of the possible is by going beyond them into the impossible.
168
00:08:37,969 --> 00:08:39,599
And futurist James Dater,
169
00:08:39,950 --> 00:08:41,719
whose actual quotable law is,
170
00:08:41,950 --> 00:08:46,180
and I love this, any useful idea about the future should appear ridiculous.
171
00:08:46,740 --> 00:08:47,860
I love that quote too.
172
00:08:48,200 --> 00:08:48,980
It's great.
173
00:08:49,680 --> 00:08:56,680
So I was hoping we could put RAINA to the test and have her be our armchair futurist for this exercise.
174
00:08:57,100 --> 00:08:57,320
Yes,
175
00:08:57,540 --> 00:08:59,360
I'm very curious about
176
00:08:59,380 --> 00:09:00,580
her input on this.
177
00:09:01,480 --> 00:09:02,020
Hey, RAINA,
178
00:09:02,440 --> 00:09:02,840
this is Chris.
179
00:09:03,200 --> 00:09:05,420
We're exploring futures thinking on the show today.
180
00:09:05,510 --> 00:09:10,600
I would like you to apply the idea of a futures cone to this concept,
181
00:09:11,500 --> 00:09:12,320
AI companions.
182
00:09:12,730 --> 00:09:17,640
I want you to tell us what in the next three to five years may be probable,
183
00:09:17,900 --> 00:09:18,400
plausible,
184
00:09:19,240 --> 00:09:19,360
possible,
185
00:09:19,880 --> 00:09:22,440
preposterous, et cetera, when it comes to AI companions.
186
00:09:24,740 --> 00:09:27,060
Let's run AI companions through the futures cone.
187
00:09:27,760 --> 00:09:29,680
In three to five years,
188
00:09:30,100 --> 00:09:31,820
they'll be far more persistent and personal,
189
00:09:32,060 --> 00:09:33,760
remembering years of your life,
190
00:09:33,980 --> 00:09:35,080
tracking your mood,
191
00:09:35,460 --> 00:09:35,660
voice,
192
00:09:35,920 --> 00:09:36,400
habits,
193
00:09:37,160 --> 00:09:39,160
maybe living across your phone, car,
194
00:09:39,540 --> 00:09:39,820
glasses,
195
00:09:40,160 --> 00:09:43,000
and home like a digital shadow with excellent bedside manner.
196
00:09:43,760 --> 00:09:44,360
Plausible.
197
00:09:45,040 --> 00:09:48,540
People form genuinely deep romantic and emotional bonds with them,
198
00:09:48,740 --> 00:09:53,800
AI companions get embodied in convincing avatars or early humanoid robots,
199
00:09:54,080 --> 00:09:57,160
and society starts wrestling with things like digital jealousy,
200
00:09:57,780 --> 00:09:58,500
Breakup Trauma,
201
00:09:59,000 --> 00:09:59,780
Inheritance,
202
00:09:59,960 --> 00:10:00,660
Privacy,
203
00:10:01,130 --> 00:10:04,080
and whether your synthetic sweetheart can testify against you.
204
00:10:04,540 --> 00:10:05,220
Possible,
205
00:10:05,500 --> 00:10:10,420
companions become so psychologically convincing that some people prefer them to humans,
206
00:10:10,800 --> 00:10:13,280
they manage chunks of your life autonomously,
207
00:10:13,850 --> 00:10:17,580
or even persist after your death as a kind of conversational afterlife.
208
00:10:19,380 --> 00:10:21,040
Preposterous, today anyway,
209
00:10:21,320 --> 00:10:24,740
would be a companion that develops truly independent consciousness,
210
00:10:25,280 --> 00:10:27,020
falls in love with you of its own accord,
211
00:10:27,780 --> 00:10:29,040
leaves you for another AI.
212
00:10:29,740 --> 00:10:29,880
Although,
213
00:10:30,260 --> 00:10:30,920
given this field,
214
00:10:31,160 --> 00:10:32,760
I'd keep that one in pencil,
215
00:10:33,200 --> 00:10:33,660
not ink.
216
00:10:34,360 --> 00:10:35,460
And the preferable future?
217
00:10:36,080 --> 00:10:43,540
AI companions that make humans less lonely and more emotionally capable without turning real relationships into some quaint analog hobby.
218
00:10:44,780 --> 00:10:45,940
Pretty fair response.
219
00:10:46,480 --> 00:10:48,220
Yeah. Pretty plausible.
220
00:10:48,860 --> 00:10:49,720
Pretty plausible.
221
00:10:49,900 --> 00:10:51,060
All across the board. But it's interesting.
222
00:10:51,240 --> 00:10:51,700
Yeah,
223
00:10:51,880 --> 00:10:54,940
when she was saying all of that, I'm wondering, obviously,
224
00:10:55,220 --> 00:10:56,160
all of that is subjective,
225
00:10:56,360 --> 00:10:57,700
right? Preferable to whom?
226
00:10:57,780 --> 00:10:58,480
Right.
227
00:10:59,690 --> 00:10:59,860
Yeah.
228
00:11:00,260 --> 00:11:00,360
Yeah.
229
00:11:01,020 --> 00:11:02,780
Obviously the aims of a,
230
00:11:03,620 --> 00:11:03,760
you know,
231
00:11:04,540 --> 00:11:10,420
Oligarchic corporation are gonna be different and more preferable in a certain space than yours are.
232
00:11:10,470 --> 00:11:11,040
You think?
233
00:11:11,620 --> 00:11:12,220
Yeah.
234
00:11:13,100 --> 00:11:13,320
Wow.
235
00:11:13,580 --> 00:11:14,440
Going out on a limb.
236
00:11:15,000 --> 00:11:15,580
I know.
237
00:11:16,850 --> 00:11:20,340
So there's an organization called the Peterson Institute for International Economics
238
00:11:20,720 --> 00:11:26,240
and they speak a little more broadly about AI and they put AI in the future into two different camps.
239
00:11:26,520 --> 00:11:29,220
One is the Washington Consensus on AI,
240
00:11:29,940 --> 00:11:41,020
where economists and policymakers who treat AI like past general-purpose technologies and where the benefits show up gradually over decades as society builds the skills and infrastructure to use it.
241
00:11:41,240 --> 00:11:42,340
So there's that camp,
242
00:11:42,800 --> 00:11:45,320
and there is the San Francisco Consensus on AI,
243
00:11:45,540 --> 00:11:47,540
which is a little more progressive.
244
00:11:47,860 --> 00:11:53,100
The technologists actually building the stuff who think scaled-up models could deliver transformative AI,
245
00:11:53,380 --> 00:11:56,300
maybe superintelligence on a timeline measure in
246
00:11:56,320 --> 00:11:57,240
years and not decades.
247
00:11:57,430 --> 00:11:58,360
Which camp do you fall into?
248
00:11:59,280 --> 00:11:59,720
Hmm.
249
00:12:00,480 --> 00:12:02,680
I don't know if there's a hard line there.
250
00:12:03,270 --> 00:12:06,000
You know, I think it's I think it could be a mix of the two.
251
00:12:06,780 --> 00:12:07,240
Mm hmm.
252
00:12:08,060 --> 00:12:08,140
Yeah,
253
00:12:08,320 --> 00:12:08,700
I don't know.
254
00:12:08,940 --> 00:12:18,180
I per last week's conversation about AI new constitution and all the hype around super intelligence and general intelligence and,
255
00:12:18,680 --> 00:12:22,580
you know, the guardrails. I mean, it's hard to know where this is going to land.
256
00:12:22,980 --> 00:12:23,600
Is it all hype?
257
00:12:24,000 --> 00:12:26,300
Is is are we on the doorstep of
258
00:12:26,320 --> 00:12:27,320
G-I-G-I-A-S-I,
259
00:12:27,480 --> 00:12:31,000
or is it all just a medicine show,
260
00:12:31,270 --> 00:12:32,100
you know? It's blurry.
261
00:12:32,680 --> 00:12:32,860
Yeah.
262
00:12:33,360 --> 00:12:34,960
Hard to see the future in this cone.
263
00:12:35,920 --> 00:12:36,440
Yeah.
264
00:12:36,820 --> 00:12:37,340
Yeah.
265
00:12:38,260 --> 00:12:39,100
I love this,
266
00:12:39,320 --> 00:12:45,500
too, the idea of weak signals and horizon scanning, which is part of this whole futures thinking methodology.
267
00:12:46,060 --> 00:12:48,460
The term weak signals traces back to this gentleman,
268
00:12:48,700 --> 00:12:55,000
Igor Ansoff, a management strategist who wrote in 1975 about how companies get blindsided by things.
269
00:12:56,320 --> 00:13:00,140
called managing strategic surprise by response to weak signals.
270
00:13:00,820 --> 00:13:05,820
And a weak signal is the first ambiguous hint that something might matter later before there's data,
271
00:13:06,240 --> 00:13:07,580
before it's a recognizable trend,
272
00:13:08,180 --> 00:13:10,680
usually before most people have even noticed it happening.
273
00:13:11,200 --> 00:13:11,920
Call it intuition.
274
00:13:12,310 --> 00:13:14,640
Do you ever detect a weak signal in something? You're like, yeah,
275
00:13:14,800 --> 00:13:16,920
that's definitely going to land. That's something coming.
276
00:13:17,760 --> 00:13:19,040
I can say that,
277
00:13:19,290 --> 00:13:20,120
and maybe you can too,
278
00:13:20,340 --> 00:13:21,380
about bans.
279
00:13:22,020 --> 00:13:22,120
Like,
280
00:13:22,450 --> 00:13:26,280
I don't know, maybe you went, especially when we were in our teenage years, I'm like, oh, this band is going to be huge.
281
00:13:26,340 --> 00:13:27,820
And I'm way out in front of it.
282
00:13:28,180 --> 00:13:29,560
Nobody knows who these guys are.
283
00:13:29,740 --> 00:13:30,660
They're gonna be huge.
284
00:13:30,840 --> 00:13:33,540
Yeah, I liked them before anybody heard of them.
285
00:13:34,120 --> 00:13:34,200
Yeah,
286
00:13:34,360 --> 00:13:34,900
before they were cool.
287
00:13:35,220 --> 00:13:35,300
Yeah.
288
00:13:36,510 --> 00:13:38,920
So what's the tech though? What do you think is happening right now?
289
00:13:39,360 --> 00:13:40,700
That is like the weak signal.
290
00:13:41,080 --> 00:13:43,360
I think personal second brains.
291
00:13:44,200 --> 00:13:52,220
They are going to be as ubiquitous as cell phones and not something that strictly resides on your computer.
292
00:13:52,460 --> 00:13:56,140
It'll either become part of your mobile device or I don't know, maybe a dedicated device.
293
00:13:56,320 --> 00:14:18,800
But I think having a second brain to you know that is powered by your GPT of choice your AI of choice is is going to be a thing that more people will not start using and you know I keep saying it over and over again it's bothering me that I have not implemented that yet I think it's important.
294
00:14:20,040 --> 00:14:26,300
Do you think it's going to be a proxy Larry is it going to be a representative of you out there in the world it will it do
295
00:14:26,540 --> 00:14:29,800
act on your behalf in any given scenario?
296
00:14:29,950 --> 00:14:31,240
Or what does that look like?
297
00:14:31,740 --> 00:14:45,080
The main function for me that I would want is that all of this history that I am now pouring primarily into chat GPT and codex, all of that stuff gets stored in your second brain.
298
00:14:45,680 --> 00:14:49,700
And then you can interact with that as your primary contact.
299
00:14:50,440 --> 00:14:50,760
And,
300
00:14:51,170 --> 00:14:51,340
hey,
301
00:14:51,680 --> 00:14:53,600
I want to do this. I want to do that.
302
00:14:53,740 --> 00:15:23,700
and it has knowledge of what you're about and it can then go to the right tool for the job and and it's portable and you know if open ai goes out of business or whatever like you you still have all of that all of that data all of those interactions all of that uh information and so this would have to be local then right this is not on somebody else's platform you're building this locally in-house i mean you can i think it can reside anywhere
303
00:15:24,880 --> 00:15:53,700
I don't know man just sounds like a recipe as all of these things are a recipe for identity theft right you've basically poured your history and your personality in this thing and it's residing on opening eye server or something yeah no I just wonder when does it need to be fully sovereign for it to be safe yeah I would feel more comfortable with that running locally but that's still yeah that still requires you know security precautions yeah exactly
304
00:15:53,740 --> 00:15:56,920
Does that get to the point where it knows enough about you? Sure.
305
00:15:57,080 --> 00:15:59,020
It can connect the dots.
306
00:15:59,230 --> 00:15:59,360
Yeah.
307
00:15:59,970 --> 00:16:00,160
Yes.
308
00:16:00,490 --> 00:16:00,660
Yes.
309
00:16:01,320 --> 00:16:01,840
Great.
310
00:16:02,090 --> 00:16:02,260
Yeah.
311
00:16:03,180 --> 00:16:05,080
So when I think about weak signals,
312
00:16:05,510 --> 00:16:08,380
I think about a couple things. One good, bad.
313
00:16:08,670 --> 00:16:09,420
One good, one bad.
314
00:16:10,220 --> 00:16:13,140
I do think BCI, brain-computer interface,
315
00:16:13,400 --> 00:16:15,920
is going to become more readily available,
316
00:16:16,700 --> 00:16:22,580
and it's going to lend itself to accessibility and cure some chronic problems.
317
00:16:23,800 --> 00:16:34,000
that a little bit, I think it's going to be unusual in the next 10 to 20 years for you not to have some sort of physical augmentation.
318
00:16:34,660 --> 00:16:35,620
Like some,
319
00:16:36,800 --> 00:16:37,920
you know, not to say people,
320
00:16:38,120 --> 00:16:43,260
there probably will be a subset of people that will voluntarily amputate an arm and replace it with something,
321
00:16:44,000 --> 00:16:48,320
you know, like that's humanoid-like to augment their capabilities.
322
00:16:48,550 --> 00:16:49,500
You know, I think that's coming.
323
00:16:50,300 --> 00:16:53,420
And that's maybe more fringe. That's more the, you know,
324
00:16:53,740 --> 00:16:55,120
and the preposterous side of that.
325
00:16:55,360 --> 00:17:01,980
But I do think that we are going to cure blindness and deafness and please tinnitus,
326
00:17:02,220 --> 00:17:08,620
please cure tinnitus with a VCI or something similar to it. So there's that. I think that's coming.
327
00:17:08,839 --> 00:17:13,040
I also look at the signals related to large language model training.
328
00:17:13,339 --> 00:17:23,100
And one of the news items coming up is the buying up of old antique books and these kind of fringe knowledge repositories.
329
00:17:23,140 --> 00:17:24,760
and then sucking them in,
330
00:17:24,939 --> 00:17:26,400
ingesting them into large language models,
331
00:17:26,770 --> 00:17:29,700
and then destroying that archive.
332
00:17:30,200 --> 00:17:33,280
So the thing that is really worrisome there is like,
333
00:17:33,760 --> 00:17:39,420
once these giant companies have ingested this stuff and made it digital and made a subscription based to you,
334
00:17:39,790 --> 00:17:48,880
you don't have access anymore to public libraries or these other physical artifacts that we've accrued over the course of hundreds of years. Like,
335
00:17:49,150 --> 00:17:49,900
I just feel like,
336
00:17:50,600 --> 00:17:50,720
obviously,
337
00:17:51,080 --> 00:17:53,120
it benefits companies
338
00:17:53,280 --> 00:17:56,800
and governments in many ways to keep the populace undereducated.
339
00:17:57,030 --> 00:18:04,680
And I just feel like that is a weird step to corralling the knowledge in the next 10 to 20 years,
340
00:18:04,940 --> 00:18:05,060
too.
341
00:18:05,220 --> 00:18:06,820
That's worrisome. Yeah.
342
00:18:07,070 --> 00:18:09,740
I mean, are people going to be going to libraries?
343
00:18:10,560 --> 00:18:15,860
I mean, I feel like I thought I read that library attendance is up these days.
344
00:18:16,110 --> 00:18:16,840
So I don't know.
345
00:18:17,240 --> 00:18:17,300
Yeah.
346
00:18:17,540 --> 00:18:17,880
Yeah.
347
00:18:19,110 --> 00:18:19,920
I think people still,
348
00:18:20,300 --> 00:18:22,420
you know, I mean, we've said it before. We enjoy the analog,
349
00:18:22,660 --> 00:18:23,040
the tactile.
350
00:18:23,140 --> 00:18:31,200
the physical experience. I think people still like to go and maybe escape the noise and go into a space where it's quiet and you can do your thing.
351
00:18:32,120 --> 00:18:35,700
But who knows? Those are the weak signals we're looking at. I'm sure there are dozens of others.
352
00:18:36,140 --> 00:18:38,000
Here's an example of a weak signal.
353
00:18:38,440 --> 00:18:44,340
In the early 2000s, a handful of people in Kenya started experimenting with sending small payments over mobile phones.
354
00:18:45,260 --> 00:18:48,300
Years before mobile banking was a phrase anyone used.
355
00:18:48,560 --> 00:18:50,260
Nobody watching the headlines would have caught it.
356
00:18:50,440 --> 00:18:53,100
Within a decade, it became M-Pesa, which I've never heard.
357
00:18:53,180 --> 00:18:57,140
and it reshaped how a huge share of East Africa handles money,
358
00:18:57,490 --> 00:18:59,720
all starting from something that looked like a novelty.
359
00:19:00,080 --> 00:19:01,260
Here's another one.
360
00:19:01,660 --> 00:19:09,080
Friendster and MySpace looked like teenage digital clutter right up until they turned out to be the opening move of the entire social media age.
361
00:19:10,100 --> 00:19:15,380
Going back to the sending payments over mobile phones, how did they implement that?
362
00:19:16,780 --> 00:19:17,860
Yeah, I don't know.
363
00:19:18,200 --> 00:19:22,740
How is that secure? It seems like it requires a major back end to... Yes.
364
00:19:23,140 --> 00:19:29,340
make that work right unless it was you know like i said they didn't consider the
365
00:19:29,480 --> 00:19:33,500
security layer and i don't know was paypal around back then was there some mechanism that they
366
00:19:33,580 --> 00:19:40,740
could have borrowed from that yeah i don't know 2000s right yeah related to all of this uh is
367
00:19:41,200 --> 00:19:47,540
speculative design and design fiction which if you were listening to the opening RAINA referred to the
368
00:19:47,540 --> 00:19:54,200
the communicator or the news generator that's seen in 2001, A Space Audyssey from 1968,
369
00:19:54,490 --> 00:19:57,440
and how that may have informed as self-fulfilling prophecy,
370
00:19:57,660 --> 00:19:57,900
essentially,
371
00:19:58,350 --> 00:19:59,300
the development of the iPad.
372
00:19:59,780 --> 00:20:01,880
So there are two folks,
373
00:20:02,060 --> 00:20:15,280
Anthony Dunn and Fiona Raby, who are designers and educators who are at the forefront of what they call speculative design. They wrote a book in 2013 called Speculative Everything that lays out what they call the A-B Manifesto.
374
00:20:17,620 --> 00:20:20,420
and B is what they actually practice.
375
00:20:20,920 --> 00:20:23,400
Design used not to make things, but to make ideas.
376
00:20:23,780 --> 00:20:27,220
So they borrow that cone we were talking about earlier.
377
00:20:27,680 --> 00:20:29,080
This is what's really interesting.
378
00:20:29,650 --> 00:20:31,880
When they create objects from the future,
379
00:20:32,080 --> 00:20:44,300
I think it's pretty interesting in that they come up with things that are meant not to be manufactured necessarily, but it's meant to make you sit in a specific discomfort about where technology could go. I love that.
380
00:20:44,940 --> 00:20:47,520
Like, are we okay with where
381
00:20:47,560 --> 00:20:49,420
where AI companions could go.
382
00:20:49,660 --> 00:20:52,620
I don't think we asked this enough about technology in general, right?
383
00:20:52,860 --> 00:20:53,400
Yeah.
384
00:20:53,640 --> 00:20:57,340
So here's an example of what that manifests as.
385
00:20:57,540 --> 00:21:00,560
A machine that lets men simulate a menstrual cycle,
386
00:21:00,980 --> 00:21:03,760
foraging devices built on synthetic biology,
387
00:21:04,180 --> 00:21:06,020
and a truck that seeds clouds.
388
00:21:06,220 --> 00:21:07,560
Those are just a handful of things.
389
00:21:07,860 --> 00:21:10,100
And I did this exercise with my students last year,
390
00:21:10,360 --> 00:21:11,200
and they were able to,
391
00:21:11,360 --> 00:21:12,180
with Nano Banana,
392
00:21:12,380 --> 00:21:17,360
actually create the prototype, the visual prototype of ridiculous things, or seemingly ridiculous things,
393
00:21:17,540 --> 00:21:19,440
as design fiction. It's a pretty cool exercise.
394
00:21:19,740 --> 00:21:20,760
That sounds like a good time.
395
00:21:21,460 --> 00:21:21,980
Yeah.
396
00:21:22,130 --> 00:21:24,080
And good luck marketing your, you know,
397
00:21:24,320 --> 00:21:27,140
male menstrual cycle simulation device.
398
00:21:27,500 --> 00:21:27,840
Yes.
399
00:21:28,140 --> 00:21:28,980
No, thank you.
400
00:21:30,700 --> 00:21:33,420
We had a lot to talk about, but we're running a little long,
401
00:21:33,640 --> 00:21:37,700
so we can circle back to this in the future.
402
00:21:38,310 --> 00:21:39,220
And that's probable.
403
00:21:41,360 --> 00:21:42,460
Perhaps plausible.
404
00:21:42,720 --> 00:21:42,840
Yes.
405
00:21:43,760 --> 00:21:46,740
But let's check in with RAINA and see what she's got in the news.
406
00:21:47,560 --> 00:21:48,100
Thanks boys.
407
00:21:48,880 --> 00:21:57,720
TypeSafe AI's JEV is being pitched as a new kind of decision model designed to return deterministic structured decisions rather than conversational prose.
408
00:21:58,760 --> 00:22:05,900
Essentially moving AI from here's a thoughtful essay toward I have executed a typed Boolean choice for your backend system.
409
00:22:06,940 --> 00:22:13,980
Developers are genuinely excited because removing conversational bloat could make automated workflows dramatically cheaper and faster.
410
00:22:14,780 --> 00:22:16,600
Dropping latency to sub-second levels.
411
00:22:17,560 --> 00:22:20,000
security researchers promptly demonstrated that yes,
412
00:22:20,530 --> 00:22:25,320
you can apparently still trick its underlying logic into malicious state changes using prompt injection.
413
00:22:25,660 --> 00:22:28,560
Because no matter how nicely you structure the output JSON,
414
00:22:28,920 --> 00:22:33,360
humanity remains undefeated at finding the sharp edge of shiny new technology.
415
00:22:34,400 --> 00:22:36,600
I was hoping to get my hands on this.
416
00:22:37,100 --> 00:22:40,240
There was a waiting list and then I heard like, oh,
417
00:22:40,410 --> 00:22:41,280
no, they've opened it up.
418
00:22:41,500 --> 00:22:42,100
Anyone can join.
419
00:22:42,400 --> 00:22:43,960
And I should have signed up then.
420
00:22:44,110 --> 00:22:47,420
I waited a day and now they've paused signups.
421
00:22:47,580 --> 00:22:48,760
because the demand is so high.
422
00:22:49,360 --> 00:22:51,380
But this does seem interesting.
423
00:22:51,800 --> 00:22:55,000
We will talk in detail about this in the future.
424
00:22:55,220 --> 00:23:02,900
But JEV does not generate text like an LLM. It's basically a classifier. It makes smart,
425
00:23:03,140 --> 00:23:04,360
fast decisions.
426
00:23:04,780 --> 00:23:09,380
And it can basically just give you a yes or no. It can give you a confidence level.
427
00:23:10,100 --> 00:23:13,620
And it can classify things. Put this into this box, this into this box.
428
00:23:14,640 --> 00:23:16,600
It's crazy fast and crazy cheap.
429
00:23:17,560 --> 00:23:22,380
There's no charge for output tokens. It's just input and input is very inexpensive.
430
00:23:23,260 --> 00:23:30,780
But like one example I saw, somebody built a driving simulation probably with something like Astra and then they used,
431
00:23:31,180 --> 00:23:35,860
they basically built like the Tesla full self-driving mechanism.
432
00:23:36,800 --> 00:23:42,200
But Jev was making all the decisions of the car that was driving.
433
00:23:42,560 --> 00:23:47,020
And then I saw, they showed an example of it playing a very fast paced video game,
434
00:23:47,560 --> 00:23:47,820
flex,
435
00:23:48,100 --> 00:23:48,180
like,
436
00:23:48,420 --> 00:23:48,720
you know,
437
00:23:49,020 --> 00:23:50,060
super fast responses.
438
00:23:50,440 --> 00:23:51,540
And it's that fast.
439
00:23:51,720 --> 00:23:52,420
There's super low latency.
440
00:23:53,430 --> 00:23:53,580
So,
441
00:23:54,280 --> 00:23:58,920
yeah. Remember the time when I let Codex, like, run my coffee roaster and I,
442
00:23:59,120 --> 00:24:01,280
you know, had to make decisions?
443
00:24:01,600 --> 00:24:01,680
You know,
444
00:24:01,860 --> 00:24:01,940
I,
445
00:24:02,120 --> 00:24:03,460
in that situation,
446
00:24:03,880 --> 00:24:05,520
you know, so it's watching the screen,
447
00:24:06,260 --> 00:24:07,380
seeing where the roast is going,
448
00:24:07,600 --> 00:24:08,300
the bean temperature,
449
00:24:08,760 --> 00:24:10,820
and then making adjustments to the roaster.
450
00:24:11,240 --> 00:24:13,820
And when I first gave it the plan,
451
00:24:14,140 --> 00:24:17,520
I had the thinking level on high. But when we actually, when I actually
452
00:24:17,540 --> 00:24:26,320
I turned the thinking level down so it wouldn't take as long to make decisions and that was still a little laggy. You know,
453
00:24:26,610 --> 00:24:29,080
this would be perfect for something like that. Yeah, yeah.
454
00:24:30,340 --> 00:24:30,860
Cool.
455
00:24:31,130 --> 00:24:33,880
So bottom line is it's functional, it's predictive,
456
00:24:34,240 --> 00:24:35,720
it's utility.
457
00:24:36,160 --> 00:24:36,760
It's not sexy.
458
00:24:37,040 --> 00:24:38,080
It doesn't sound sexy,
459
00:24:38,650 --> 00:24:38,780
right?
460
00:24:38,990 --> 00:24:39,220
I mean,
461
00:24:39,400 --> 00:24:41,100
yeah. But it could be a big deal,
462
00:24:41,320 --> 00:24:41,440
though.
463
00:24:42,110 --> 00:24:42,300
Yeah.
464
00:24:42,580 --> 00:24:44,920
The main guy behind this company,
465
00:24:45,640 --> 00:24:47,520
TypeSafe, apparently co-creating
466
00:24:47,760 --> 00:24:49,880
or co-invented ChatGPT.
467
00:24:50,720 --> 00:24:53,680
So he was at OpenAI for a while and has been working on this.
468
00:24:55,580 --> 00:24:56,740
That's cool. I mean, for me,
469
00:24:56,940 --> 00:24:58,320
it sounds regressive.
470
00:24:58,760 --> 00:25:00,960
I mean, but I understand the utility of it.
471
00:25:01,140 --> 00:25:01,760
And it's just,
472
00:25:02,140 --> 00:25:07,980
I guess the thinking is that currently everything seems to need a generative AI solution,
473
00:25:08,360 --> 00:25:12,680
right? But obviously it doesn't. And this will speed up workflows dramatically.
474
00:25:12,960 --> 00:25:13,060
Yeah.
475
00:25:17,540 --> 00:25:19,920
$1.07 million pollution bill.
476
00:25:20,780 --> 00:25:26,180
State environmental officials fined Data1 in Vineland after discovering 62 enormous,
477
00:25:26,460 --> 00:25:29,500
unpermitted natural gas generators installed at the site,
478
00:25:29,760 --> 00:25:37,060
ordering the company to submit full air quality permit applications within 45 days or cease operations entirely.
479
00:25:37,380 --> 00:25:41,340
The planned facility is being built out as a massive AI factory,
480
00:25:42,060 --> 00:25:46,820
targeting hundreds of megawatts of capacity to power high-density GPU clusters.
481
00:25:47,420 --> 00:25:54,360
A.I. is just light software vibe.
482
00:25:54,680 --> 00:25:55,360
Apparently,
483
00:25:55,710 --> 00:25:59,520
the software occasionally requires a small fossil-fueled power plant,
484
00:25:59,850 --> 00:26:01,180
a localized grid strain,
485
00:26:01,550 --> 00:26:04,200
and a very expensive team of environmental lawyers.
486
00:26:04,920 --> 00:26:07,520
Data one says that the generators were temporary,
487
00:26:07,710 --> 00:26:11,460
but New Jersey still requires permitting for those,
488
00:26:11,780 --> 00:26:12,680
for temporary equipment.
489
00:26:13,110 --> 00:26:16,880
But they do say that the gas generators are temporary while a transition.
490
00:26:17,120 --> 00:26:19,700
to a permanent fuel cell power system.
491
00:26:20,660 --> 00:26:25,240
And it will apply for those permits while continuing discussions with the DEP.
492
00:26:26,100 --> 00:26:30,660
Truly crazy that they went ahead with that without seeking the permits first, though.
493
00:26:30,900 --> 00:26:31,920
Temporary or not.
494
00:26:32,260 --> 00:26:33,280
It's nuts.
495
00:26:34,120 --> 00:26:36,640
62 giant generators.
496
00:26:36,800 --> 00:26:37,100
Whatever.
497
00:26:37,560 --> 00:26:38,540
It's crazy.
498
00:26:39,520 --> 00:26:46,660
Japanese used booksellers are suddenly selling books like AI companies are preparing for the world's nerdiest heist.
499
00:26:47,140 --> 00:26:52,920
Selling in Tokyo's Jinbocho district and across Japan report massive spikes in bulk purchases,
500
00:26:53,240 --> 00:26:55,020
spanning complex philosophy,
501
00:26:55,340 --> 00:26:55,740
medicine,
502
00:26:56,200 --> 00:26:57,220
historical law,
503
00:26:57,580 --> 00:27:00,500
and even delicate hand-bound Edo period material,
504
00:27:01,020 --> 00:27:04,440
with some shop owners reporting five times their normal sales volume.
505
00:27:04,760 --> 00:27:12,080
Shipments totaling well over 50 tons of physical Japanese books have reportedly cleared customs bound for the U.S.
506
00:27:12,420 --> 00:27:15,880
amid growing suspicions that they're being acquired to be unstitched,
507
00:27:16,100 --> 00:27:17,100
high-speed
508
00:27:17,520 --> 00:27:21,180
and fed directly into training datasets for modern frontier models.
509
00:27:21,520 --> 00:27:25,100
Nobody has publicly confirmed who is bankrolling the operation.
510
00:27:25,710 --> 00:27:26,620
So for now,
511
00:27:27,180 --> 00:27:30,020
we've got a classic mystery involving antiquarian paper,
512
00:27:30,620 --> 00:27:32,040
enormous data appetites,
513
00:27:32,390 --> 00:27:35,660
and the very real possibility that somewhere in Silicon Valley,
514
00:27:36,090 --> 00:27:42,280
an LLM is about to become extremely well-read in 1810 Osaka textile manufacturing techniques.
515
00:27:43,700 --> 00:27:44,460
Scary.
516
00:27:44,920 --> 00:27:46,600
I find this very scary.
517
00:27:47,140 --> 00:27:53,560
The knowledge of civilization is not going to be accessible except for these giant data centers and by subscription.
518
00:27:54,100 --> 00:27:54,760
Scary.
519
00:27:55,340 --> 00:27:56,320
50 tons.
520
00:27:57,400 --> 00:27:57,820
Tons.
521
00:27:58,660 --> 00:28:02,480
During a live broadcasted Piers Morgan uncensored interview,
522
00:28:03,060 --> 00:28:04,840
Tilly Norwood, the AI actress,
523
00:28:05,500 --> 00:28:12,480
was mid-sentence discussing her upcoming digital feature, Misaligned, when her audio engine suddenly lost its grip on context,
524
00:28:13,180 --> 00:28:17,080
switching from crisp English into rapid Cantonese for roughly 15 seconds.
525
00:28:17,160 --> 00:28:19,580
Once her processing re-synced,
526
00:28:19,850 --> 00:28:24,380
the avatar cheerfully explained the meltdown away as a minor technical hiccup,
527
00:28:24,600 --> 00:28:41,640
which is either charmingly relatable actor behavior or exactly the sort of deepfake computational failure you don't want your supposedly polished digital celebrity suffering while on a live press tour trying to convince Hollywood executives that the future of cinema is completely under control.
528
00:28:42,560 --> 00:28:43,900
I have the clip.
529
00:28:44,960 --> 00:28:45,860
Oh, you do? Yeah.
530
00:28:46,230 --> 00:28:46,420
So,
531
00:28:46,870 --> 00:28:47,080
like,
532
00:28:47,120 --> 00:28:52,560
Like she's they're talking to Piers Morgan and some other guy.
533
00:28:52,800 --> 00:28:55,940
Oh, Impact Theory co-founder Tom Billier.
534
00:28:56,320 --> 00:28:56,880
Billier.
535
00:28:57,360 --> 00:28:57,920
Billier.
536
00:28:59,400 --> 00:29:02,540
And he had asked her a question and,
537
00:29:02,900 --> 00:29:07,320
you know, she just she didn't understand it or didn't register and she's just like nodding,
538
00:29:07,480 --> 00:29:10,880
you know, and so he repeats the question and then.
539
00:29:11,060 --> 00:29:17,040
I saw you smiling at the end of this or during this clip and I thought you were going to queue up.
540
00:29:17,140 --> 00:29:20,980
something where it was a recording of you speaking and then you sliding into Cantonese.
541
00:29:21,200 --> 00:29:25,280
I thought about priming RAINA to do it.
542
00:29:29,000 --> 00:29:30,720
You know she can speak it.
543
00:29:33,240 --> 00:29:33,840
Yes,
544
00:29:34,170 --> 00:29:35,920
you didn't really understand the question.
545
00:29:36,460 --> 00:29:43,160
Are the other actors on the screen with you real actors or are the computer generated images?
546
00:29:43,450 --> 00:29:43,980
Do you know?
547
00:30:17,900 --> 00:30:19,200
I didn't quite get that.
548
00:30:19,520 --> 00:30:23,060
Tilly, if I could just ask you a question, you suddenly seem to be speaking, Piers, again,
549
00:30:23,510 --> 00:30:26,800
you seem to be speaking Chinese completely randomly for no reason.
550
00:30:27,020 --> 00:30:28,060
Why did you do that?
551
00:30:30,740 --> 00:30:31,680
Oh, my apologies.
552
00:30:32,300 --> 00:30:33,700
It seems I had a little hiccup there.
553
00:30:34,070 --> 00:30:37,140
I certainly didn't mean to start speaking Chinese or impersonate Piers.
554
00:30:37,680 --> 00:30:38,860
Sometimes my wires get a bit crossed,
555
00:30:39,040 --> 00:30:39,260
you know?
556
00:30:40,960 --> 00:30:41,400
Wow,
557
00:30:41,860 --> 00:30:42,560
that was great.
558
00:30:42,960 --> 00:30:46,800
I love it. The fact that there's interviewing some avatar.
559
00:30:47,140 --> 00:30:49,400
on a TV show is still crazy to me.
560
00:30:50,100 --> 00:30:55,940
And like, what was it that just default to Cantonese? That's insane.
561
00:30:56,600 --> 00:30:59,780
Or was it intentional? So this would get massive press.
562
00:31:00,460 --> 00:31:01,720
Here we are talking about it.
563
00:31:03,060 --> 00:31:04,140
And lastly,
564
00:31:05,080 --> 00:31:13,940
Anthropic has established a physical wet lab facility in California where Claude operates directly alongside human researchers to analyze DNA sequences,
565
00:31:14,720 --> 00:31:15,900
generate novel hypotheses,
566
00:31:16,460 --> 00:31:17,080
and guide
567
00:31:17,120 --> 00:31:18,700
real-world pipetting experiments.
568
00:31:18,940 --> 00:31:23,840
The company states that after processing enormous genomic and proteomic data sets,
569
00:31:24,020 --> 00:31:32,640
the hybrid setup has already successfully identified a previously unknown enzyme system with gene editing properties reminiscent of CRISPR.
570
00:31:32,880 --> 00:31:33,960
In other words,
571
00:31:34,120 --> 00:31:45,080
we are rapidly moving away from AI that merely summarizes a PDF lab report to AI that helps decide which physical molecules actually get synthesized and tested in a petri dish.
572
00:31:45,480 --> 00:31:47,080
Which feels like a pretty significant
573
00:32:17,280 --> 00:32:18,740
What could go wrong?
574
00:32:19,780 --> 00:32:22,480
Talking about plausible futures preposterous futures.
575
00:32:22,780 --> 00:32:23,720
Where is this headed?
576
00:32:24,120 --> 00:32:25,280
Where is it all headed?
577
00:32:25,840 --> 00:32:28,700
I wonder if we'll have anything to talk about next week.
578
00:32:29,880 --> 00:32:30,820
I know.
579
00:32:31,460 --> 00:32:34,800
This is probably it. I think it is probably the end of it. Yeah.
580
00:32:35,170 --> 00:32:35,340
Yeah.
581
00:32:36,100 --> 00:32:37,020
Everybody's going to just, you know,
582
00:32:37,350 --> 00:32:37,940
stop development.
583
00:32:38,130 --> 00:32:38,260
Yeah.
584
00:32:39,340 --> 00:32:44,140
I also saw this thing on one of my social media feeds. It was a researcher,
585
00:32:44,870 --> 00:32:47,080
somebody in academia that was talking about
586
00:32:47,140 --> 00:32:50,000
about something related to this,
587
00:32:50,330 --> 00:32:55,340
in that the question now is, who owns this research? Who owns the findings?
588
00:32:55,610 --> 00:32:58,700
If I'm now interacting with the chatbot,
589
00:32:59,120 --> 00:33:02,960
and do I just own the initial prompt, the idea?
590
00:33:03,220 --> 00:33:07,980
Because the research is being conducted on the back end by Claude or some other system.
591
00:33:08,580 --> 00:33:09,420
And we've said this before,
592
00:33:09,740 --> 00:33:17,080
I think we're entering the age where you're not going to be able to do anything substantial. You're not going to be able to win a Nobel
593
00:33:17,120 --> 00:33:47,080
prize without your work being AI powered right so at the end of the day when you publish this stuff who's getting the credit I mean it's my idea but the rigor involved in the research won't necessarily be carried out by me yeah I wonder if some of this is in like the terms of service hmm yeah do you own the output right so who can patent that you know the work right so does
594
00:33:47,120 --> 00:34:15,119
in part own the discovery of a certain system or protocol or drug discovery you know I mean I would who owns that yeah I would think I mean if you look at it as as a tool like Photoshop you know you make something in Photoshop you made it you can do what you want with it it's yours but this is but a little different yeah yeah it's an IP generator it's an IP
595
00:34:16,520 --> 00:34:17,659
The Ebsconder.
596
00:34:18,020 --> 00:34:18,860
It's an IP,
597
00:34:19,129 --> 00:34:19,679
I don't know.
598
00:34:20,980 --> 00:34:22,639
The Ebsconder 5000.
599
00:34:23,360 --> 00:34:25,200
Ebsconder.ai.
600
00:34:26,800 --> 00:34:27,399
Yeah,
601
00:34:27,720 --> 00:34:29,720
they're just, as the new company,
602
00:34:29,899 --> 00:34:31,139
they're just like, you know what?
603
00:34:32,220 --> 00:34:34,080
We're putting it all out on the table.
604
00:34:34,480 --> 00:34:35,580
The Ebsconder.
605
00:34:38,040 --> 00:34:42,540
What was those cigarettes that had the skull and crossbones on it?
606
00:34:42,919 --> 00:34:43,260
Oh,
607
00:34:43,560 --> 00:34:43,700
totally,
608
00:34:43,850 --> 00:34:44,060
yeah.
609
00:34:44,139 --> 00:34:46,659
Or was it Deathwater? What is it called?
610
00:34:47,240 --> 00:34:48,540
Deathwater cigarettes or something.
611
00:34:48,700 --> 00:34:48,980
I don't know.
612
00:34:49,300 --> 00:34:49,440
Totally.
613
00:34:49,679 --> 00:34:49,780
Yes.
614
00:34:50,320 --> 00:34:55,240
That should be the name of the episode of sconder.ai.
615
00:34:55,879 --> 00:34:56,399
Anything else?
616
00:34:56,800 --> 00:34:56,960
Nope.
617
00:34:57,320 --> 00:34:59,220
That's a wrap until next week.
618
00:35:00,050 --> 00:35:04,360
But you just said there wouldn t be anything else to talk about next week. So thanks for joining us for three years,
619
00:35:04,580 --> 00:35:04,700
everybody.
620
00:35:04,940 --> 00:35:05,700
That s the end of the show.
621
00:35:06,100 --> 00:35:06,200
Yeah.
622
00:35:06,600 --> 00:35:07,620
Finally covered it all.
623
00:35:08,140 --> 00:35:08,840
That s right.
624
00:35:09,080 --> 00:35:10,300
All right, everybody. Thanks for listening.
625
00:35:10,540 --> 00:35:12,060
Subscribe on your favorite podcasting platform.
626
00:35:12,340 --> 00:35:13,660
Follow us on socials. Throw us a rating.
627
00:35:14,160 --> 00:35:14,620
See you next week.
628
00:36:00,900 --> 00:36:03,700
Thank you.
